4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/namei.h>
35 #include "delegation.h"
37 #define NFS_PARANOIA 1
38 /* #define NFS_DEBUG_VERBOSE 1 */
40 static int nfs_opendir(struct inode *, struct file *);
41 static int nfs_readdir(struct file *, void *, filldir_t);
42 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
43 static int nfs_cached_lookup(struct inode *, struct dentry *,
44 struct nfs_fh *, struct nfs_fattr *);
45 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
46 static int nfs_mkdir(struct inode *, struct dentry *, int);
47 static int nfs_rmdir(struct inode *, struct dentry *);
48 static int nfs_unlink(struct inode *, struct dentry *);
49 static int nfs_symlink(struct inode *, struct dentry *, const char *);
50 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
51 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
52 static int nfs_rename(struct inode *, struct dentry *,
53 struct inode *, struct dentry *);
54 static int nfs_fsync_dir(struct file *, struct dentry *, int);
56 struct file_operations nfs_dir_operations = {
57 .read = generic_read_dir,
58 .readdir = nfs_readdir,
60 .release = nfs_release,
61 .fsync = nfs_fsync_dir,
64 struct inode_operations nfs_dir_inode_operations = {
69 .symlink = nfs_symlink,
74 .permission = nfs_permission,
75 .getattr = nfs_getattr,
76 .setattr = nfs_setattr,
81 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
82 struct inode_operations nfs4_dir_inode_operations = {
84 .lookup = nfs_atomic_lookup,
87 .symlink = nfs_symlink,
92 .permission = nfs_permission,
93 .getattr = nfs_getattr,
94 .setattr = nfs_setattr,
97 #endif /* CONFIG_NFS_V4 */
103 nfs_opendir(struct inode *inode, struct file *filp)
108 /* Call generic open code in order to cache credentials */
110 res = nfs_open(inode, filp);
115 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
119 unsigned long page_index;
122 struct nfs_entry *entry;
123 decode_dirent_t decode;
126 } nfs_readdir_descriptor_t;
128 /* Now we cache directories properly, by stuffing the dirent
129 * data directly in the page cache.
131 * Inode invalidation due to refresh etc. takes care of
132 * _everything_, no sloppy entry flushing logic, no extraneous
133 * copying, network direct to page cache, the way it was meant
136 * NOTE: Dirent information verification is done always by the
137 * page-in of the RPC reply, nowhere else, this simplies
138 * things substantially.
141 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
143 struct file *file = desc->file;
144 struct inode *inode = file->f_dentry->d_inode;
145 struct rpc_cred *cred = nfs_file_cred(file);
146 unsigned long timestamp;
149 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
153 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
154 NFS_SERVER(inode)->dtsize, desc->plus);
156 /* We requested READDIRPLUS, but the server doesn't grok it */
157 if (error == -ENOTSUPP && desc->plus) {
158 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
159 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
165 SetPageUptodate(page);
166 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
167 /* Ensure consistent page alignment of the data.
168 * Note: assumes we have exclusive access to this mapping either
169 * throught inode->i_sem or some other mechanism.
171 if (page->index == 0) {
172 invalidate_inode_pages(inode->i_mapping);
173 NFS_I(inode)->readdir_timestamp = timestamp;
180 nfs_zap_caches(inode);
186 int dir_decode(nfs_readdir_descriptor_t *desc)
189 p = desc->decode(p, desc->entry, desc->plus);
197 void dir_page_release(nfs_readdir_descriptor_t *desc)
200 page_cache_release(desc->page);
206 * Given a pointer to a buffer that has already been filled by a call
207 * to readdir, find the next entry.
209 * If the end of the buffer has been reached, return -EAGAIN, if not,
210 * return the offset within the buffer of the next entry to be
214 int find_dirent(nfs_readdir_descriptor_t *desc, struct page *page)
216 struct nfs_entry *entry = desc->entry;
220 while((status = dir_decode(desc)) == 0) {
221 dfprintk(VFS, "NFS: found cookie %Lu\n", (long long)entry->cookie);
222 if (entry->prev_cookie == desc->target)
224 if (loop_count++ > 200) {
229 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
234 * Find the given page, and call find_dirent() in order to try to
235 * return the next entry.
238 int find_dirent_page(nfs_readdir_descriptor_t *desc)
240 struct inode *inode = desc->file->f_dentry->d_inode;
244 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
246 page = read_cache_page(inode->i_mapping, desc->page_index,
247 (filler_t *)nfs_readdir_filler, desc);
249 status = PTR_ERR(page);
252 if (!PageUptodate(page))
255 /* NOTE: Someone else may have changed the READDIRPLUS flag */
257 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
258 status = find_dirent(desc, page);
260 dir_page_release(desc);
262 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
265 page_cache_release(page);
270 * Recurse through the page cache pages, and return a
271 * filled nfs_entry structure of the next directory entry if possible.
273 * The target for the search is 'desc->target'.
276 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
281 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (long long)desc->target);
283 res = find_dirent_page(desc);
286 /* Align to beginning of next page */
288 if (loop_count++ > 200) {
293 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
297 static unsigned int nfs_type2dtype[] = {
310 unsigned int nfs_type_to_d_type(enum nfs_ftype type)
312 return nfs_type2dtype[type];
316 * Once we've found the start of the dirent within a page: fill 'er up...
319 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
322 struct file *file = desc->file;
323 struct nfs_entry *entry = desc->entry;
324 unsigned long fileid;
328 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)desc->target);
331 unsigned d_type = DT_UNKNOWN;
332 /* Note: entry->prev_cookie contains the cookie for
333 * retrieving the current dirent on the server */
334 fileid = nfs_fileid_to_ino_t(entry->ino);
336 /* Use readdirplus info */
337 if (desc->plus && (entry->fattr->valid & NFS_ATTR_FATTR))
338 d_type = nfs_type_to_d_type(entry->fattr->type);
340 res = filldir(dirent, entry->name, entry->len,
341 entry->prev_cookie, fileid, d_type);
344 file->f_pos = desc->target = entry->cookie;
345 if (dir_decode(desc) != 0) {
349 if (loop_count++ > 200) {
354 dir_page_release(desc);
356 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (long long)desc->target, res);
361 * If we cannot find a cookie in our cache, we suspect that this is
362 * because it points to a deleted file, so we ask the server to return
363 * whatever it thinks is the next entry. We then feed this to filldir.
364 * If all goes well, we should then be able to find our way round the
365 * cache on the next call to readdir_search_pagecache();
367 * NOTE: we cannot add the anonymous page to the pagecache because
368 * the data it contains might not be page aligned. Besides,
369 * we should already have a complete representation of the
370 * directory in the page cache by the time we get here.
373 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
376 struct file *file = desc->file;
377 struct inode *inode = file->f_dentry->d_inode;
378 struct rpc_cred *cred = nfs_file_cred(file);
379 struct page *page = NULL;
382 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (long long)desc->target);
384 page = alloc_page(GFP_HIGHUSER);
389 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->target,
391 NFS_SERVER(inode)->dtsize,
393 NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
395 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
396 if (desc->error >= 0) {
397 if ((status = dir_decode(desc)) == 0)
398 desc->entry->prev_cookie = desc->target;
404 status = nfs_do_filldir(desc, dirent, filldir);
406 /* Reset read descriptor so it searches the page cache from
407 * the start upon the next call to readdir_search_pagecache() */
408 desc->page_index = 0;
409 desc->entry->cookie = desc->entry->prev_cookie = 0;
410 desc->entry->eof = 0;
412 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
415 dir_page_release(desc);
419 /* The file offset position is now represented as a true offset into the
420 * page cache as is the case in most of the other filesystems.
422 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
424 struct dentry *dentry = filp->f_dentry;
425 struct inode *inode = dentry->d_inode;
426 nfs_readdir_descriptor_t my_desc,
428 struct nfs_entry my_entry;
430 struct nfs_fattr fattr;
435 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
442 * filp->f_pos points to the file offset in the page cache.
443 * but if the cache has meanwhile been zapped, we need to
444 * read from the last dirent to revalidate f_pos
447 memset(desc, 0, sizeof(*desc));
450 desc->target = filp->f_pos;
451 desc->decode = NFS_PROTO(inode)->decode_dirent;
452 desc->plus = NFS_USE_READDIRPLUS(inode);
454 my_entry.cookie = my_entry.prev_cookie = 0;
457 my_entry.fattr = &fattr;
458 desc->entry = &my_entry;
460 while(!desc->entry->eof) {
461 res = readdir_search_pagecache(desc);
462 if (res == -EBADCOOKIE) {
463 /* This means either end of directory */
464 if (desc->entry->cookie != desc->target) {
465 /* Or that the server has 'lost' a cookie */
466 res = uncached_readdir(desc, dirent, filldir);
473 if (res == -ETOOSMALL && desc->plus) {
474 NFS_FLAGS(inode) &= ~NFS_INO_ADVISE_RDPLUS;
475 nfs_zap_caches(inode);
477 desc->entry->eof = 0;
483 res = nfs_do_filldir(desc, dirent, filldir);
498 * All directory operations under NFS are synchronous, so fsync()
499 * is a dummy operation.
501 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
507 * A check for whether or not the parent directory has changed.
508 * In the case it has, we assume that the dentries are untrustworthy
509 * and may need to be looked up again.
511 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
515 if ((NFS_FLAGS(dir) & NFS_INO_INVALID_ATTR) != 0
516 || nfs_attribute_timeout(dir))
518 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
521 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
523 dentry->d_fsdata = (void *)verf;
527 * Whenever an NFS operation succeeds, we know that the dentry
528 * is valid, so we update the revalidation timestamp.
530 static inline void nfs_renew_times(struct dentry * dentry)
532 dentry->d_time = jiffies;
536 int nfs_lookup_verify_inode(struct inode *inode, int isopen)
538 struct nfs_server *server = NFS_SERVER(inode);
540 if (isopen && !(server->flags & NFS_MOUNT_NOCTO))
541 return __nfs_revalidate_inode(server, inode);
542 return nfs_revalidate_inode(server, inode);
546 * We judge how long we want to trust negative
547 * dentries by looking at the parent inode mtime.
549 * If parent mtime has changed, we revalidate, else we wait for a
550 * period corresponding to the parent's attribute cache timeout value.
553 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
554 struct nameidata *nd)
560 /* Don't revalidate a negative dentry if we're creating a new file */
561 if ((ndflags & LOOKUP_CREATE) && !(ndflags & LOOKUP_CONTINUE))
563 return !nfs_check_verifier(dir, dentry);
567 * This is called every time the dcache has a lookup hit,
568 * and we should check whether we can really trust that
571 * NOTE! The hit can be a negative hit too, don't assume
574 * If the parent directory is seen to have changed, we throw out the
575 * cached dentry and do a new lookup.
577 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
581 struct dentry *parent;
583 struct nfs_fh fhandle;
584 struct nfs_fattr fattr;
585 unsigned long verifier;
588 parent = dget_parent(dentry);
590 dir = parent->d_inode;
591 inode = dentry->d_inode;
593 if (nd && !(nd->flags & LOOKUP_CONTINUE) && (nd->flags & LOOKUP_OPEN))
597 if (nfs_neg_need_reval(dir, dentry, nd))
602 if (is_bad_inode(inode)) {
603 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
604 dentry->d_parent->d_name.name, dentry->d_name.name);
608 /* Revalidate parent directory attribute cache */
609 nfs_revalidate_inode(NFS_SERVER(dir), dir);
611 /* Force a full look up iff the parent directory has changed */
612 if (nfs_check_verifier(dir, dentry)) {
613 if (nfs_lookup_verify_inode(inode, isopen))
619 * Note: we're not holding inode->i_sem and so may be racing with
620 * operations that change the directory. We therefore save the
621 * change attribute *before* we do the RPC call.
623 verifier = nfs_save_change_attribute(dir);
624 error = nfs_cached_lookup(dir, dentry, &fhandle, &fattr);
626 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
628 if (nfs_lookup_verify_inode(inode, isopen))
630 goto out_valid_renew;
633 if (NFS_STALE(inode))
636 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
639 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
641 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
645 nfs_renew_times(dentry);
646 nfs_set_verifier(dentry, verifier);
655 if (inode && S_ISDIR(inode->i_mode)) {
656 /* Purge readdir caches. */
657 nfs_zap_caches(inode);
658 /* If we have submounts, don't unhash ! */
659 if (have_submounts(dentry))
661 shrink_dcache_parent(dentry);
670 * This is called from dput() when d_count is going to 0.
672 static int nfs_dentry_delete(struct dentry *dentry)
674 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
675 dentry->d_parent->d_name.name, dentry->d_name.name,
678 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
679 /* Unhash it, so that ->d_iput() would be called */
682 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
683 /* Unhash it, so that ancestors of killed async unlink
684 * files will be cleaned up during umount */
692 * Called when the dentry loses inode.
693 * We use it to clean up silly-renamed files.
695 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
697 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
700 nfs_complete_unlink(dentry);
703 /* When creating a negative dentry, we want to renew d_time */
704 nfs_renew_times(dentry);
708 struct dentry_operations nfs_dentry_operations = {
709 .d_revalidate = nfs_lookup_revalidate,
710 .d_delete = nfs_dentry_delete,
711 .d_iput = nfs_dentry_iput,
715 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
717 if (NFS_PROTO(dir)->version == 2)
719 if (!nd || (nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_CREATE))
721 return (nd->intent.open.flags & O_EXCL) != 0;
724 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
726 struct inode *inode = NULL;
728 struct nfs_fh fhandle;
729 struct nfs_fattr fattr;
731 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
732 dentry->d_parent->d_name.name, dentry->d_name.name);
734 error = -ENAMETOOLONG;
735 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
739 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
742 /* Revalidate parent directory attribute cache */
743 nfs_revalidate_inode(NFS_SERVER(dir), dir);
745 /* If we're doing an exclusive create, optimize away the lookup */
746 if (nfs_is_exclusive_create(dir, nd))
749 error = nfs_cached_lookup(dir, dentry, &fhandle, &fattr);
751 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name,
753 if (error == -ENOENT)
759 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
764 d_add(dentry, inode);
765 nfs_renew_times(dentry);
766 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
771 return ERR_PTR(error);
775 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
777 struct dentry_operations nfs4_dentry_operations = {
778 .d_revalidate = nfs_open_revalidate,
779 .d_delete = nfs_dentry_delete,
780 .d_iput = nfs_dentry_iput,
783 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
787 /* Check that we are indeed trying to open this file */
788 if ((nd->flags & LOOKUP_CONTINUE) || !(nd->flags & LOOKUP_OPEN))
790 /* NFS does not (yet) have a stateful open for directories */
791 if (nd->flags & LOOKUP_DIRECTORY)
793 /* Are we trying to write to a read only partition? */
794 if ((IS_RDONLY(dir) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
795 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
800 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
802 struct inode *inode = NULL;
805 /* Check that we are indeed trying to open this file */
806 if (!is_atomic_open(dir, nd))
809 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
810 error = -ENAMETOOLONG;
813 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
815 /* Let vfs_create() deal with O_EXCL */
816 if (nd->intent.open.flags & O_EXCL)
819 /* Open the file on the server */
821 /* Revalidate parent directory attribute cache */
822 nfs_revalidate_inode(NFS_SERVER(dir), dir);
824 if (nd->intent.open.flags & O_CREAT) {
825 nfs_begin_data_update(dir);
826 inode = nfs4_atomic_open(dir, dentry, nd);
827 nfs_end_data_update(dir);
829 inode = nfs4_atomic_open(dir, dentry, nd);
832 error = PTR_ERR(inode);
834 /* Make a negative dentry */
838 /* This turned out not to be a regular file */
840 if (!(nd->intent.open.flags & O_NOFOLLOW))
849 d_add(dentry, inode);
850 nfs_renew_times(dentry);
851 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
854 return ERR_PTR(error);
856 return nfs_lookup(dir, dentry, nd);
859 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
861 struct dentry *parent = NULL;
862 struct inode *inode = dentry->d_inode;
864 unsigned long verifier;
865 int openflags, ret = 0;
867 parent = dget_parent(dentry);
868 dir = parent->d_inode;
869 if (!is_atomic_open(dir, nd))
871 /* We can't create new files in nfs_open_revalidate(), so we
872 * optimize away revalidation of negative dentries.
876 /* NFS only supports OPEN on regular files */
877 if (!S_ISREG(inode->i_mode))
879 openflags = nd->intent.open.flags;
880 /* We cannot do exclusive creation on a positive dentry */
881 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
883 /* We can't create new files, or truncate existing ones here */
884 openflags &= ~(O_CREAT|O_TRUNC);
887 * Note: we're not holding inode->i_sem and so may be racing with
888 * operations that change the directory. We therefore save the
889 * change attribute *before* we do the RPC call.
892 verifier = nfs_save_change_attribute(dir);
893 ret = nfs4_open_revalidate(dir, dentry, openflags);
895 nfs_set_verifier(dentry, verifier);
904 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
906 return nfs_lookup_revalidate(dentry, nd);
908 #endif /* CONFIG_NFSV4 */
911 int find_dirent_name(nfs_readdir_descriptor_t *desc, struct page *page, struct dentry *dentry)
913 struct nfs_entry *entry = desc->entry;
916 while((status = dir_decode(desc)) == 0) {
917 if (entry->len != dentry->d_name.len)
919 if (memcmp(entry->name, dentry->d_name.name, entry->len))
921 if (!(entry->fattr->valid & NFS_ATTR_FATTR))
929 * Use the cached Readdirplus results in order to avoid a LOOKUP call
930 * whenever we believe that the parent directory has not changed.
932 * We assume that any file creation/rename changes the directory mtime.
933 * As this results in a page cache invalidation whenever it occurs,
934 * we don't require any other tests for cache coherency.
937 int nfs_cached_lookup(struct inode *dir, struct dentry *dentry,
938 struct nfs_fh *fh, struct nfs_fattr *fattr)
940 nfs_readdir_descriptor_t desc;
941 struct nfs_server *server;
942 struct nfs_entry entry;
944 unsigned long timestamp;
947 if (!NFS_USE_READDIRPLUS(dir))
949 server = NFS_SERVER(dir);
950 /* Don't use readdirplus unless the cache is stable */
951 if ((server->flags & NFS_MOUNT_NOAC) != 0
952 || nfs_caches_unstable(dir)
953 || nfs_attribute_timeout(dir))
955 if ((NFS_FLAGS(dir) & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) != 0)
957 timestamp = NFS_I(dir)->readdir_timestamp;
962 desc.decode = NFS_PROTO(dir)->decode_dirent;
967 for(;(page = find_get_page(dir->i_mapping, desc.page_index)); desc.page_index++) {
970 if (PageUptodate(page)) {
971 void * kaddr = kmap_atomic(page, KM_USER0);
973 res = find_dirent_name(&desc, page, dentry);
974 kunmap_atomic(kaddr, KM_USER0);
976 page_cache_release(page);
985 fattr->timestamp = timestamp;
990 * Code common to create, mkdir, and mknod.
992 static int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
993 struct nfs_fattr *fattr)
998 /* We may have been initialized further down */
1001 if (fhandle->size == 0) {
1002 struct inode *dir = dentry->d_parent->d_inode;
1003 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1007 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1008 struct nfs_server *server = NFS_SB(dentry->d_sb);
1009 error = server->rpc_ops->getattr(server, fhandle, fattr);
1013 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1015 d_instantiate(dentry, inode);
1016 nfs_renew_times(dentry);
1017 nfs_set_verifier(dentry, nfs_save_change_attribute(dentry->d_parent->d_inode));
1027 * Following a failed create operation, we drop the dentry rather
1028 * than retain a negative dentry. This avoids a problem in the event
1029 * that the operation succeeded on the server, but an error in the
1030 * reply path made it appear to have failed.
1032 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1033 struct nameidata *nd)
1036 struct inode *inode;
1040 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1041 dir->i_ino, dentry->d_name.name);
1043 attr.ia_mode = mode;
1044 attr.ia_valid = ATTR_MODE;
1046 if (nd && (nd->flags & LOOKUP_CREATE))
1047 open_flags = nd->intent.open.flags;
1050 * The 0 argument passed into the create function should one day
1051 * contain the O_EXCL flag if requested. This allows NFSv3 to
1052 * select the appropriate create strategy. Currently open_namei
1053 * does not pass the create flags.
1056 nfs_begin_data_update(dir);
1057 inode = NFS_PROTO(dir)->create(dir, &dentry->d_name, &attr, open_flags);
1058 nfs_end_data_update(dir);
1059 if (!IS_ERR(inode)) {
1060 d_instantiate(dentry, inode);
1061 nfs_renew_times(dentry);
1062 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1065 error = PTR_ERR(inode);
1073 * See comments for nfs_proc_create regarding failed operations.
1076 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1079 struct nfs_fattr fattr;
1080 struct nfs_fh fhandle;
1083 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1084 dir->i_ino, dentry->d_name.name);
1086 if (!new_valid_dev(rdev))
1089 attr.ia_mode = mode;
1090 attr.ia_valid = ATTR_MODE;
1093 nfs_begin_data_update(dir);
1094 error = NFS_PROTO(dir)->mknod(dir, &dentry->d_name, &attr, rdev,
1096 nfs_end_data_update(dir);
1098 error = nfs_instantiate(dentry, &fhandle, &fattr);
1106 * See comments for nfs_proc_create regarding failed operations.
1108 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1111 struct nfs_fattr fattr;
1112 struct nfs_fh fhandle;
1115 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1116 dir->i_ino, dentry->d_name.name);
1118 attr.ia_valid = ATTR_MODE;
1119 attr.ia_mode = mode | S_IFDIR;
1124 * Always drop the dentry, we can't always depend on
1125 * the fattr returned by the server (AIX seems to be
1126 * broken). We're better off doing another lookup than
1127 * depending on potentially bogus information.
1131 nfs_begin_data_update(dir);
1132 error = NFS_PROTO(dir)->mkdir(dir, &dentry->d_name, &attr, &fhandle,
1134 nfs_end_data_update(dir);
1136 error = nfs_instantiate(dentry, &fhandle, &fattr);
1143 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1147 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1148 dir->i_ino, dentry->d_name.name);
1151 nfs_begin_data_update(dir);
1152 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1153 /* Ensure the VFS deletes this inode */
1154 if (error == 0 && dentry->d_inode != NULL)
1155 dentry->d_inode->i_nlink = 0;
1156 nfs_end_data_update(dir);
1162 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1164 static unsigned int sillycounter;
1165 const int i_inosize = sizeof(dir->i_ino)*2;
1166 const int countersize = sizeof(sillycounter)*2;
1167 const int slen = strlen(".nfs") + i_inosize + countersize;
1170 struct dentry *sdentry;
1173 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1174 dentry->d_parent->d_name.name, dentry->d_name.name,
1175 atomic_read(&dentry->d_count));
1178 if (!dentry->d_inode)
1179 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1180 dentry->d_parent->d_name.name, dentry->d_name.name);
1183 * We don't allow a dentry to be silly-renamed twice.
1186 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1189 sprintf(silly, ".nfs%*.*lx",
1190 i_inosize, i_inosize, dentry->d_inode->i_ino);
1194 char *suffix = silly + slen - countersize;
1198 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1200 dfprintk(VFS, "trying to rename %s to %s\n",
1201 dentry->d_name.name, silly);
1203 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1205 * N.B. Better to return EBUSY here ... it could be
1206 * dangerous to delete the file while it's in use.
1208 if (IS_ERR(sdentry))
1210 } while(sdentry->d_inode != NULL); /* need negative lookup */
1212 qsilly.name = silly;
1213 qsilly.len = strlen(silly);
1214 nfs_begin_data_update(dir);
1215 if (dentry->d_inode) {
1216 nfs_begin_data_update(dentry->d_inode);
1217 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1219 nfs_end_data_update(dentry->d_inode);
1221 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1223 nfs_end_data_update(dir);
1225 nfs_renew_times(dentry);
1226 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1227 d_move(dentry, sdentry);
1228 error = nfs_async_unlink(dentry);
1229 /* If we return 0 we don't unlink */
1237 * Remove a file after making sure there are no pending writes,
1238 * and after checking that the file has only one user.
1240 * We invalidate the attribute cache and free the inode prior to the operation
1241 * to avoid possible races if the server reuses the inode.
1243 static int nfs_safe_remove(struct dentry *dentry)
1245 struct inode *dir = dentry->d_parent->d_inode;
1246 struct inode *inode = dentry->d_inode;
1249 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1250 dentry->d_parent->d_name.name, dentry->d_name.name);
1252 /* If the dentry was sillyrenamed, we simply call d_delete() */
1253 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1258 nfs_begin_data_update(dir);
1259 if (inode != NULL) {
1260 nfs_begin_data_update(inode);
1261 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1262 /* The VFS may want to delete this inode */
1265 nfs_end_data_update(inode);
1267 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1268 nfs_end_data_update(dir);
1273 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1274 * belongs to an active ".nfs..." file and we return -EBUSY.
1276 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1278 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1281 int need_rehash = 0;
1283 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1284 dir->i_ino, dentry->d_name.name);
1287 spin_lock(&dcache_lock);
1288 spin_lock(&dentry->d_lock);
1289 if (atomic_read(&dentry->d_count) > 1) {
1290 spin_unlock(&dentry->d_lock);
1291 spin_unlock(&dcache_lock);
1292 error = nfs_sillyrename(dir, dentry);
1296 if (!d_unhashed(dentry)) {
1300 spin_unlock(&dentry->d_lock);
1301 spin_unlock(&dcache_lock);
1302 error = nfs_safe_remove(dentry);
1304 nfs_renew_times(dentry);
1305 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1306 } else if (need_rehash)
1313 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1316 struct nfs_fattr sym_attr;
1317 struct nfs_fh sym_fh;
1318 struct qstr qsymname;
1321 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1322 dir->i_ino, dentry->d_name.name, symname);
1325 if (dentry->d_inode)
1326 printk("nfs_proc_symlink: %s/%s not negative!\n",
1327 dentry->d_parent->d_name.name, dentry->d_name.name);
1330 * Fill in the sattr for the call.
1331 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1333 attr.ia_valid = ATTR_MODE;
1334 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1336 qsymname.name = symname;
1337 qsymname.len = strlen(symname);
1340 nfs_begin_data_update(dir);
1341 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1342 &attr, &sym_fh, &sym_attr);
1343 nfs_end_data_update(dir);
1345 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1347 if (error == -EEXIST)
1348 printk("nfs_proc_symlink: %s/%s already exists??\n",
1349 dentry->d_parent->d_name.name, dentry->d_name.name);
1357 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1359 struct inode *inode = old_dentry->d_inode;
1362 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1363 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1364 dentry->d_parent->d_name.name, dentry->d_name.name);
1367 * Drop the dentry in advance to force a new lookup.
1368 * Since nfs_proc_link doesn't return a file handle,
1369 * we can't use the existing dentry.
1374 nfs_begin_data_update(dir);
1375 nfs_begin_data_update(inode);
1376 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1377 nfs_end_data_update(inode);
1378 nfs_end_data_update(dir);
1385 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1386 * different file handle for the same inode after a rename (e.g. when
1387 * moving to a different directory). A fail-safe method to do so would
1388 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1389 * rename the old file using the sillyrename stuff. This way, the original
1390 * file in old_dir will go away when the last process iput()s the inode.
1394 * It actually works quite well. One needs to have the possibility for
1395 * at least one ".nfs..." file in each directory the file ever gets
1396 * moved or linked to which happens automagically with the new
1397 * implementation that only depends on the dcache stuff instead of
1398 * using the inode layer
1400 * Unfortunately, things are a little more complicated than indicated
1401 * above. For a cross-directory move, we want to make sure we can get
1402 * rid of the old inode after the operation. This means there must be
1403 * no pending writes (if it's a file), and the use count must be 1.
1404 * If these conditions are met, we can drop the dentries before doing
1407 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1408 struct inode *new_dir, struct dentry *new_dentry)
1410 struct inode *old_inode = old_dentry->d_inode;
1411 struct inode *new_inode = new_dentry->d_inode;
1412 struct dentry *dentry = NULL, *rehash = NULL;
1416 * To prevent any new references to the target during the rename,
1417 * we unhash the dentry and free the inode in advance.
1420 if (!d_unhashed(new_dentry)) {
1422 rehash = new_dentry;
1425 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1426 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1427 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1428 atomic_read(&new_dentry->d_count));
1431 * First check whether the target is busy ... we can't
1432 * safely do _any_ rename if the target is in use.
1434 * For files, make a copy of the dentry and then do a
1435 * silly-rename. If the silly-rename succeeds, the
1436 * copied dentry is hashed and becomes the new target.
1440 if (S_ISDIR(new_inode->i_mode))
1442 else if (atomic_read(&new_dentry->d_count) > 2) {
1444 /* copy the target dentry's name */
1445 dentry = d_alloc(new_dentry->d_parent,
1446 &new_dentry->d_name);
1450 /* silly-rename the existing target ... */
1451 err = nfs_sillyrename(new_dir, new_dentry);
1453 new_dentry = rehash = dentry;
1455 /* instantiate the replacement target */
1456 d_instantiate(new_dentry, NULL);
1457 } else if (atomic_read(&new_dentry->d_count) > 1) {
1458 /* dentry still busy? */
1460 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1461 new_dentry->d_parent->d_name.name,
1462 new_dentry->d_name.name,
1463 atomic_read(&new_dentry->d_count));
1471 * ... prune child dentries and writebacks if needed.
1473 if (atomic_read(&old_dentry->d_count) > 1) {
1474 nfs_wb_all(old_inode);
1475 shrink_dcache_parent(old_dentry);
1479 d_delete(new_dentry);
1481 nfs_begin_data_update(old_dir);
1482 nfs_begin_data_update(new_dir);
1483 nfs_begin_data_update(old_inode);
1484 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1485 new_dir, &new_dentry->d_name);
1486 nfs_end_data_update(old_inode);
1487 nfs_end_data_update(new_dir);
1488 nfs_end_data_update(old_dir);
1493 if (!S_ISDIR(old_inode->i_mode))
1494 d_move(old_dentry, new_dentry);
1495 nfs_renew_times(new_dentry);
1496 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1499 /* new dentry created? */
1506 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1508 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1510 if (cache->cred != cred
1511 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1512 || (NFS_FLAGS(inode) & NFS_INO_INVALID_ATTR))
1514 memcpy(res, cache, sizeof(*res));
1518 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1520 struct nfs_access_entry *cache = &NFS_I(inode)->cache_access;
1522 if (cache->cred != set->cred) {
1524 put_rpccred(cache->cred);
1525 cache->cred = get_rpccred(set->cred);
1527 cache->jiffies = set->jiffies;
1528 cache->mask = set->mask;
1531 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1533 struct nfs_access_entry cache;
1536 status = nfs_access_get_cached(inode, cred, &cache);
1540 /* Be clever: ask server to check for all possible rights */
1541 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1543 cache.jiffies = jiffies;
1544 status = NFS_PROTO(inode)->access(inode, &cache);
1547 nfs_access_add_cache(inode, &cache);
1549 if ((cache.mask & mask) == mask)
1554 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1556 struct rpc_cred *cred;
1557 int mode = inode->i_mode;
1562 if (mask & MAY_WRITE) {
1565 * Nobody gets write access to a read-only fs.
1568 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
1569 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1574 * Nobody gets write access to an immutable file.
1577 if (IS_IMMUTABLE(inode))
1580 /* Are we checking permissions on anything other than lookup/execute? */
1581 if ((mask & MAY_EXEC) == 0) {
1582 /* We only need to check permissions on file open() and access() */
1583 if (!nd || !(nd->flags & (LOOKUP_OPEN|LOOKUP_ACCESS)))
1585 /* NFSv4 has atomic_open... */
1586 if (NFS_PROTO(inode)->version > 3 && (nd->flags & LOOKUP_OPEN))
1592 if (!NFS_PROTO(inode)->access)
1595 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1596 res = nfs_do_access(inode, cred, mask);
1601 nfs_revalidate_inode(NFS_SERVER(inode), inode);
1602 res = generic_permission(inode, mask, NULL);
1609 * version-control: t
1610 * kept-new-versions: 5