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/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
38 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode *, struct file *);
46 static int nfs_readdir(struct file *, void *, filldir_t);
47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49 static int nfs_mkdir(struct inode *, struct dentry *, int);
50 static int nfs_rmdir(struct inode *, struct dentry *);
51 static int nfs_unlink(struct inode *, struct dentry *);
52 static int nfs_symlink(struct inode *, struct dentry *, const char *);
53 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55 static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57 static int nfs_fsync_dir(struct file *, int);
58 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 const struct file_operations nfs_dir_operations = {
61 .llseek = nfs_llseek_dir,
62 .read = generic_read_dir,
63 .readdir = nfs_readdir,
65 .release = nfs_release,
66 .fsync = nfs_fsync_dir,
69 const struct inode_operations nfs_dir_inode_operations = {
74 .symlink = nfs_symlink,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
85 const struct inode_operations nfs3_dir_inode_operations = {
90 .symlink = nfs_symlink,
95 .permission = nfs_permission,
96 .getattr = nfs_getattr,
97 .setattr = nfs_setattr,
98 .listxattr = nfs3_listxattr,
99 .getxattr = nfs3_getxattr,
100 .setxattr = nfs3_setxattr,
101 .removexattr = nfs3_removexattr,
103 #endif /* CONFIG_NFS_V3 */
107 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
108 const struct inode_operations nfs4_dir_inode_operations = {
109 .create = nfs_create,
110 .lookup = nfs_atomic_lookup,
112 .unlink = nfs_unlink,
113 .symlink = nfs_symlink,
117 .rename = nfs_rename,
118 .permission = nfs_permission,
119 .getattr = nfs_getattr,
120 .setattr = nfs_setattr,
121 .getxattr = nfs4_getxattr,
122 .setxattr = nfs4_setxattr,
123 .listxattr = nfs4_listxattr,
126 #endif /* CONFIG_NFS_V4 */
132 nfs_opendir(struct inode *inode, struct file *filp)
136 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
137 filp->f_path.dentry->d_parent->d_name.name,
138 filp->f_path.dentry->d_name.name);
140 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
142 /* Call generic open code in order to cache credentials */
143 res = nfs_open(inode, filp);
144 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
145 /* This is a mountpoint, so d_revalidate will never
146 * have been called, so we need to refresh the
147 * inode (for close-open consistency) ourselves.
149 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
154 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
158 unsigned long page_index;
161 loff_t current_index;
162 struct nfs_entry *entry;
163 decode_dirent_t decode;
165 unsigned long timestamp;
166 unsigned long gencount;
168 } nfs_readdir_descriptor_t;
170 /* Now we cache directories properly, by stuffing the dirent
171 * data directly in the page cache.
173 * Inode invalidation due to refresh etc. takes care of
174 * _everything_, no sloppy entry flushing logic, no extraneous
175 * copying, network direct to page cache, the way it was meant
178 * NOTE: Dirent information verification is done always by the
179 * page-in of the RPC reply, nowhere else, this simplies
180 * things substantially.
183 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
185 struct file *file = desc->file;
186 struct inode *inode = file->f_path.dentry->d_inode;
187 struct rpc_cred *cred = nfs_file_cred(file);
188 unsigned long timestamp, gencount;
191 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
192 __func__, (long long)desc->entry->cookie,
197 gencount = nfs_inc_attr_generation_counter();
198 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
199 NFS_SERVER(inode)->dtsize, desc->plus);
201 /* We requested READDIRPLUS, but the server doesn't grok it */
202 if (error == -ENOTSUPP && desc->plus) {
203 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
204 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
210 desc->timestamp = timestamp;
211 desc->gencount = gencount;
212 desc->timestamp_valid = 1;
213 SetPageUptodate(page);
214 /* Ensure consistent page alignment of the data.
215 * Note: assumes we have exclusive access to this mapping either
216 * through inode->i_mutex or some other mechanism.
218 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
219 /* Should never happen */
220 nfs_zap_mapping(inode, inode->i_mapping);
230 int dir_decode(nfs_readdir_descriptor_t *desc)
232 __be32 *p = desc->ptr;
233 p = desc->decode(p, desc->entry, desc->plus);
237 if (desc->timestamp_valid) {
238 desc->entry->fattr->time_start = desc->timestamp;
239 desc->entry->fattr->gencount = desc->gencount;
241 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
246 void dir_page_release(nfs_readdir_descriptor_t *desc)
249 page_cache_release(desc->page);
255 * Given a pointer to a buffer that has already been filled by a call
256 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
258 * If the end of the buffer has been reached, return -EAGAIN, if not,
259 * return the offset within the buffer of the next entry to be
263 int find_dirent(nfs_readdir_descriptor_t *desc)
265 struct nfs_entry *entry = desc->entry;
269 while((status = dir_decode(desc)) == 0) {
270 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
271 __func__, (unsigned long long)entry->cookie);
272 if (entry->prev_cookie == *desc->dir_cookie)
274 if (loop_count++ > 200) {
283 * Given a pointer to a buffer that has already been filled by a call
284 * to readdir, find the entry at offset 'desc->file->f_pos'.
286 * If the end of the buffer has been reached, return -EAGAIN, if not,
287 * return the offset within the buffer of the next entry to be
291 int find_dirent_index(nfs_readdir_descriptor_t *desc)
293 struct nfs_entry *entry = desc->entry;
298 status = dir_decode(desc);
302 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
303 (unsigned long long)entry->cookie, desc->current_index);
305 if (desc->file->f_pos == desc->current_index) {
306 *desc->dir_cookie = entry->cookie;
309 desc->current_index++;
310 if (loop_count++ > 200) {
319 * Find the given page, and call find_dirent() or find_dirent_index in
320 * order to try to return the next entry.
323 int find_dirent_page(nfs_readdir_descriptor_t *desc)
325 struct inode *inode = desc->file->f_path.dentry->d_inode;
329 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
330 __func__, desc->page_index,
331 (long long) *desc->dir_cookie);
333 /* If we find the page in the page_cache, we cannot be sure
334 * how fresh the data is, so we will ignore readdir_plus attributes.
336 desc->timestamp_valid = 0;
337 page = read_cache_page(inode->i_mapping, desc->page_index,
338 (filler_t *)nfs_readdir_filler, desc);
340 status = PTR_ERR(page);
344 /* NOTE: Someone else may have changed the READDIRPLUS flag */
346 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
347 if (*desc->dir_cookie != 0)
348 status = find_dirent(desc);
350 status = find_dirent_index(desc);
352 dir_page_release(desc);
354 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
359 * Recurse through the page cache pages, and return a
360 * filled nfs_entry structure of the next directory entry if possible.
362 * The target for the search is '*desc->dir_cookie' if non-0,
363 * 'desc->file->f_pos' otherwise
366 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
371 /* Always search-by-index from the beginning of the cache */
372 if (*desc->dir_cookie == 0) {
373 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
374 (long long)desc->file->f_pos);
375 desc->page_index = 0;
376 desc->entry->cookie = desc->entry->prev_cookie = 0;
377 desc->entry->eof = 0;
378 desc->current_index = 0;
380 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
381 (unsigned long long)*desc->dir_cookie);
384 res = find_dirent_page(desc);
387 /* Align to beginning of next page */
389 if (loop_count++ > 200) {
395 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
399 static inline unsigned int dt_type(struct inode *inode)
401 return (inode->i_mode >> 12) & 15;
404 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
407 * Once we've found the start of the dirent within a page: fill 'er up...
410 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
413 struct file *file = desc->file;
414 struct nfs_entry *entry = desc->entry;
415 struct dentry *dentry = NULL;
420 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
421 (unsigned long long)entry->cookie);
424 unsigned d_type = DT_UNKNOWN;
425 /* Note: entry->prev_cookie contains the cookie for
426 * retrieving the current dirent on the server */
429 /* Get a dentry if we have one */
432 dentry = nfs_readdir_lookup(desc);
434 /* Use readdirplus info */
435 if (dentry != NULL && dentry->d_inode != NULL) {
436 d_type = dt_type(dentry->d_inode);
437 fileid = NFS_FILEID(dentry->d_inode);
440 res = filldir(dirent, entry->name, entry->len,
441 file->f_pos, nfs_compat_user_ino64(fileid),
446 *desc->dir_cookie = entry->cookie;
447 if (dir_decode(desc) != 0) {
451 if (loop_count++ > 200) {
456 dir_page_release(desc);
459 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
460 (unsigned long long)*desc->dir_cookie, res);
465 * If we cannot find a cookie in our cache, we suspect that this is
466 * because it points to a deleted file, so we ask the server to return
467 * whatever it thinks is the next entry. We then feed this to filldir.
468 * If all goes well, we should then be able to find our way round the
469 * cache on the next call to readdir_search_pagecache();
471 * NOTE: we cannot add the anonymous page to the pagecache because
472 * the data it contains might not be page aligned. Besides,
473 * we should already have a complete representation of the
474 * directory in the page cache by the time we get here.
477 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
480 struct file *file = desc->file;
481 struct inode *inode = file->f_path.dentry->d_inode;
482 struct rpc_cred *cred = nfs_file_cred(file);
483 struct page *page = NULL;
485 unsigned long timestamp, gencount;
487 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
488 (unsigned long long)*desc->dir_cookie);
490 page = alloc_page(GFP_HIGHUSER);
496 gencount = nfs_inc_attr_generation_counter();
497 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
498 *desc->dir_cookie, page,
499 NFS_SERVER(inode)->dtsize,
502 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
504 desc->timestamp = timestamp;
505 desc->gencount = gencount;
506 desc->timestamp_valid = 1;
507 if ((status = dir_decode(desc)) == 0)
508 desc->entry->prev_cookie = *desc->dir_cookie;
514 status = nfs_do_filldir(desc, dirent, filldir);
516 /* Reset read descriptor so it searches the page cache from
517 * the start upon the next call to readdir_search_pagecache() */
518 desc->page_index = 0;
519 desc->entry->cookie = desc->entry->prev_cookie = 0;
520 desc->entry->eof = 0;
522 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
526 dir_page_release(desc);
530 /* The file offset position represents the dirent entry number. A
531 last cookie cache takes care of the common case of reading the
534 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
536 struct dentry *dentry = filp->f_path.dentry;
537 struct inode *inode = dentry->d_inode;
538 nfs_readdir_descriptor_t my_desc,
540 struct nfs_entry my_entry;
543 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
544 dentry->d_parent->d_name.name, dentry->d_name.name,
545 (long long)filp->f_pos);
546 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
549 * filp->f_pos points to the dirent entry number.
550 * *desc->dir_cookie has the cookie for the next entry. We have
551 * to either find the entry with the appropriate number or
552 * revalidate the cookie.
554 memset(desc, 0, sizeof(*desc));
557 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
558 desc->decode = NFS_PROTO(inode)->decode_dirent;
559 desc->plus = NFS_USE_READDIRPLUS(inode);
561 my_entry.cookie = my_entry.prev_cookie = 0;
563 my_entry.fh = nfs_alloc_fhandle();
564 my_entry.fattr = nfs_alloc_fattr();
565 if (my_entry.fh == NULL || my_entry.fattr == NULL)
566 goto out_alloc_failed;
568 desc->entry = &my_entry;
570 nfs_block_sillyrename(dentry);
571 res = nfs_revalidate_mapping(inode, filp->f_mapping);
575 while(!desc->entry->eof) {
576 res = readdir_search_pagecache(desc);
578 if (res == -EBADCOOKIE) {
579 /* This means either end of directory */
580 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
581 /* Or that the server has 'lost' a cookie */
582 res = uncached_readdir(desc, dirent, filldir);
589 if (res == -ETOOSMALL && desc->plus) {
590 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
591 nfs_zap_caches(inode);
593 desc->entry->eof = 0;
599 res = nfs_do_filldir(desc, dirent, filldir);
606 nfs_unblock_sillyrename(dentry);
610 nfs_free_fattr(my_entry.fattr);
611 nfs_free_fhandle(my_entry.fh);
612 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
613 dentry->d_parent->d_name.name, dentry->d_name.name,
618 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
620 struct dentry *dentry = filp->f_path.dentry;
621 struct inode *inode = dentry->d_inode;
623 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
624 dentry->d_parent->d_name.name,
628 mutex_lock(&inode->i_mutex);
631 offset += filp->f_pos;
639 if (offset != filp->f_pos) {
640 filp->f_pos = offset;
641 nfs_file_open_context(filp)->dir_cookie = 0;
644 mutex_unlock(&inode->i_mutex);
649 * All directory operations under NFS are synchronous, so fsync()
650 * is a dummy operation.
652 static int nfs_fsync_dir(struct file *filp, int datasync)
654 struct dentry *dentry = filp->f_path.dentry;
656 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
657 dentry->d_parent->d_name.name, dentry->d_name.name,
660 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
665 * nfs_force_lookup_revalidate - Mark the directory as having changed
666 * @dir - pointer to directory inode
668 * This forces the revalidation code in nfs_lookup_revalidate() to do a
669 * full lookup on all child dentries of 'dir' whenever a change occurs
670 * on the server that might have invalidated our dcache.
672 * The caller should be holding dir->i_lock
674 void nfs_force_lookup_revalidate(struct inode *dir)
676 NFS_I(dir)->cache_change_attribute++;
680 * A check for whether or not the parent directory has changed.
681 * In the case it has, we assume that the dentries are untrustworthy
682 * and may need to be looked up again.
684 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
688 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
690 if (!nfs_verify_change_attribute(dir, dentry->d_time))
692 /* Revalidate nfsi->cache_change_attribute before we declare a match */
693 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
695 if (!nfs_verify_change_attribute(dir, dentry->d_time))
701 * Return the intent data that applies to this particular path component
703 * Note that the current set of intents only apply to the very last
704 * component of the path.
705 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
707 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
709 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
711 return nd->flags & mask;
715 * Use intent information to check whether or not we're going to do
716 * an O_EXCL create using this path component.
718 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
720 if (NFS_PROTO(dir)->version == 2)
722 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
726 * Inode and filehandle revalidation for lookups.
728 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
729 * or if the intent information indicates that we're about to open this
730 * particular file and the "nocto" mount flag is not set.
734 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
736 struct nfs_server *server = NFS_SERVER(inode);
738 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
741 /* VFS wants an on-the-wire revalidation */
742 if (nd->flags & LOOKUP_REVAL)
744 /* This is an open(2) */
745 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
746 !(server->flags & NFS_MOUNT_NOCTO) &&
747 (S_ISREG(inode->i_mode) ||
748 S_ISDIR(inode->i_mode)))
752 return nfs_revalidate_inode(server, inode);
754 return __nfs_revalidate_inode(server, inode);
758 * We judge how long we want to trust negative
759 * dentries by looking at the parent inode mtime.
761 * If parent mtime has changed, we revalidate, else we wait for a
762 * period corresponding to the parent's attribute cache timeout value.
765 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
766 struct nameidata *nd)
768 /* Don't revalidate a negative dentry if we're creating a new file */
769 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
771 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
773 return !nfs_check_verifier(dir, dentry);
777 * This is called every time the dcache has a lookup hit,
778 * and we should check whether we can really trust that
781 * NOTE! The hit can be a negative hit too, don't assume
784 * If the parent directory is seen to have changed, we throw out the
785 * cached dentry and do a new lookup.
787 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
791 struct dentry *parent;
792 struct nfs_fh *fhandle = NULL;
793 struct nfs_fattr *fattr = NULL;
796 parent = dget_parent(dentry);
797 dir = parent->d_inode;
798 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
799 inode = dentry->d_inode;
802 if (nfs_neg_need_reval(dir, dentry, nd))
807 if (is_bad_inode(inode)) {
808 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
809 __func__, dentry->d_parent->d_name.name,
810 dentry->d_name.name);
814 if (nfs_have_delegation(inode, FMODE_READ))
815 goto out_set_verifier;
817 /* Force a full look up iff the parent directory has changed */
818 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
819 if (nfs_lookup_verify_inode(inode, nd))
824 if (NFS_STALE(inode))
828 fhandle = nfs_alloc_fhandle();
829 fattr = nfs_alloc_fattr();
830 if (fhandle == NULL || fattr == NULL)
833 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
836 if (nfs_compare_fh(NFS_FH(inode), fhandle))
838 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
841 nfs_free_fattr(fattr);
842 nfs_free_fhandle(fhandle);
844 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
847 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
848 __func__, dentry->d_parent->d_name.name,
849 dentry->d_name.name);
854 nfs_mark_for_revalidate(dir);
855 if (inode && S_ISDIR(inode->i_mode)) {
856 /* Purge readdir caches. */
857 nfs_zap_caches(inode);
858 /* If we have submounts, don't unhash ! */
859 if (have_submounts(dentry))
861 if (dentry->d_flags & DCACHE_DISCONNECTED)
863 shrink_dcache_parent(dentry);
866 nfs_free_fattr(fattr);
867 nfs_free_fhandle(fhandle);
869 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
870 __func__, dentry->d_parent->d_name.name,
871 dentry->d_name.name);
874 nfs_free_fattr(fattr);
875 nfs_free_fhandle(fhandle);
877 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
878 __func__, dentry->d_parent->d_name.name,
879 dentry->d_name.name, error);
884 * This is called from dput() when d_count is going to 0.
886 static int nfs_dentry_delete(struct dentry *dentry)
888 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
889 dentry->d_parent->d_name.name, dentry->d_name.name,
892 /* Unhash any dentry with a stale inode */
893 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
896 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
897 /* Unhash it, so that ->d_iput() would be called */
900 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
901 /* Unhash it, so that ancestors of killed async unlink
902 * files will be cleaned up during umount */
909 static void nfs_drop_nlink(struct inode *inode)
911 spin_lock(&inode->i_lock);
912 if (inode->i_nlink > 0)
914 spin_unlock(&inode->i_lock);
918 * Called when the dentry loses inode.
919 * We use it to clean up silly-renamed files.
921 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
923 if (S_ISDIR(inode->i_mode))
924 /* drop any readdir cache as it could easily be old */
925 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
927 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
929 nfs_complete_unlink(dentry, inode);
934 const struct dentry_operations nfs_dentry_operations = {
935 .d_revalidate = nfs_lookup_revalidate,
936 .d_delete = nfs_dentry_delete,
937 .d_iput = nfs_dentry_iput,
940 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
943 struct dentry *parent;
944 struct inode *inode = NULL;
945 struct nfs_fh *fhandle = NULL;
946 struct nfs_fattr *fattr = NULL;
949 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
950 dentry->d_parent->d_name.name, dentry->d_name.name);
951 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
953 res = ERR_PTR(-ENAMETOOLONG);
954 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
957 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
960 * If we're doing an exclusive create, optimize away the lookup
961 * but don't hash the dentry.
963 if (nfs_is_exclusive_create(dir, nd)) {
964 d_instantiate(dentry, NULL);
969 res = ERR_PTR(-ENOMEM);
970 fhandle = nfs_alloc_fhandle();
971 fattr = nfs_alloc_fattr();
972 if (fhandle == NULL || fattr == NULL)
975 parent = dentry->d_parent;
976 /* Protect against concurrent sillydeletes */
977 nfs_block_sillyrename(parent);
978 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
979 if (error == -ENOENT)
982 res = ERR_PTR(error);
983 goto out_unblock_sillyrename;
985 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
986 res = (struct dentry *)inode;
988 goto out_unblock_sillyrename;
991 res = d_materialise_unique(dentry, inode);
994 goto out_unblock_sillyrename;
997 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
998 out_unblock_sillyrename:
999 nfs_unblock_sillyrename(parent);
1001 nfs_free_fattr(fattr);
1002 nfs_free_fhandle(fhandle);
1006 #ifdef CONFIG_NFS_V4
1007 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1009 const struct dentry_operations nfs4_dentry_operations = {
1010 .d_revalidate = nfs_open_revalidate,
1011 .d_delete = nfs_dentry_delete,
1012 .d_iput = nfs_dentry_iput,
1016 * Use intent information to determine whether we need to substitute
1017 * the NFSv4-style stateful OPEN for the LOOKUP call
1019 static int is_atomic_open(struct nameidata *nd)
1021 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1023 /* NFS does not (yet) have a stateful open for directories */
1024 if (nd->flags & LOOKUP_DIRECTORY)
1026 /* Are we trying to write to a read only partition? */
1027 if (__mnt_is_readonly(nd->path.mnt) &&
1028 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1033 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1035 struct path path = {
1036 .mnt = nd->path.mnt,
1039 struct nfs_open_context *ctx;
1040 struct rpc_cred *cred;
1041 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1043 cred = rpc_lookup_cred();
1045 return ERR_CAST(cred);
1046 ctx = alloc_nfs_open_context(&path, cred, fmode);
1049 return ERR_PTR(-ENOMEM);
1053 static int do_open(struct inode *inode, struct file *filp)
1055 nfs_fscache_set_inode_cookie(inode, filp);
1059 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1064 /* If the open_intent is for execute, we have an extra check to make */
1065 if (ctx->mode & FMODE_EXEC) {
1066 ret = nfs_may_open(ctx->path.dentry->d_inode,
1068 nd->intent.open.flags);
1072 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1074 ret = PTR_ERR(filp);
1076 nfs_file_set_open_context(filp, ctx);
1078 put_nfs_open_context(ctx);
1082 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1084 struct nfs_open_context *ctx;
1086 struct dentry *res = NULL;
1087 struct inode *inode;
1090 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1091 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1093 /* Check that we are indeed trying to open this file */
1094 if (!is_atomic_open(nd))
1097 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1098 res = ERR_PTR(-ENAMETOOLONG);
1101 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1103 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1105 if (nd->flags & LOOKUP_EXCL) {
1106 d_instantiate(dentry, NULL);
1110 ctx = nameidata_to_nfs_open_context(dentry, nd);
1111 res = ERR_CAST(ctx);
1115 open_flags = nd->intent.open.flags;
1116 if (nd->flags & LOOKUP_CREATE) {
1117 attr.ia_mode = nd->intent.open.create_mode;
1118 attr.ia_valid = ATTR_MODE;
1119 if (!IS_POSIXACL(dir))
1120 attr.ia_mode &= ~current_umask();
1122 open_flags &= ~O_EXCL;
1124 BUG_ON(open_flags & O_CREAT);
1127 /* Open the file on the server */
1128 nfs_block_sillyrename(dentry->d_parent);
1129 inode = nfs4_atomic_open(dir, ctx, open_flags, &attr);
1130 if (IS_ERR(inode)) {
1131 nfs_unblock_sillyrename(dentry->d_parent);
1132 put_nfs_open_context(ctx);
1133 switch (PTR_ERR(inode)) {
1134 /* Make a negative dentry */
1136 d_add(dentry, NULL);
1139 /* This turned out not to be a regular file */
1144 if (!(nd->intent.open.flags & O_NOFOLLOW))
1148 res = ERR_CAST(inode);
1152 res = d_add_unique(dentry, inode);
1154 dput(ctx->path.dentry);
1155 ctx->path.dentry = dget(res);
1158 nfs_intent_set_file(nd, ctx);
1159 nfs_unblock_sillyrename(dentry->d_parent);
1161 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1164 return nfs_lookup(dir, dentry, nd);
1167 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1169 struct dentry *parent = NULL;
1170 struct inode *inode = dentry->d_inode;
1172 int openflags, ret = 0;
1174 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1176 parent = dget_parent(dentry);
1177 dir = parent->d_inode;
1178 /* We can't create new files in nfs_open_revalidate(), so we
1179 * optimize away revalidation of negative dentries.
1181 if (inode == NULL) {
1182 if (!nfs_neg_need_reval(dir, dentry, nd))
1187 /* NFS only supports OPEN on regular files */
1188 if (!S_ISREG(inode->i_mode))
1190 openflags = nd->intent.open.flags;
1191 /* We cannot do exclusive creation on a positive dentry */
1192 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1194 /* We can't create new files, or truncate existing ones here */
1195 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1198 * Note: we're not holding inode->i_mutex and so may be racing with
1199 * operations that change the directory. We therefore save the
1200 * change attribute *before* we do the RPC call.
1202 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1211 return nfs_lookup_revalidate(dentry, nd);
1213 #endif /* CONFIG_NFSV4 */
1215 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1217 struct dentry *parent = desc->file->f_path.dentry;
1218 struct inode *dir = parent->d_inode;
1219 struct nfs_entry *entry = desc->entry;
1220 struct dentry *dentry, *alias;
1221 struct qstr name = {
1222 .name = entry->name,
1225 struct inode *inode;
1226 unsigned long verf = nfs_save_change_attribute(dir);
1230 if (name.name[0] == '.' && name.name[1] == '.')
1231 return dget_parent(parent);
1234 if (name.name[0] == '.')
1235 return dget(parent);
1238 spin_lock(&dir->i_lock);
1239 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1240 spin_unlock(&dir->i_lock);
1243 spin_unlock(&dir->i_lock);
1245 name.hash = full_name_hash(name.name, name.len);
1246 dentry = d_lookup(parent, &name);
1247 if (dentry != NULL) {
1248 /* Is this a positive dentry that matches the readdir info? */
1249 if (dentry->d_inode != NULL &&
1250 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1251 d_mountpoint(dentry))) {
1252 if (!desc->plus || entry->fh->size == 0)
1254 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1258 /* No, so d_drop to allow one to be created */
1262 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1264 if (name.len > NFS_SERVER(dir)->namelen)
1266 /* Note: caller is already holding the dir->i_mutex! */
1267 dentry = d_alloc(parent, &name);
1270 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1271 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1272 if (IS_ERR(inode)) {
1277 alias = d_materialise_unique(dentry, inode);
1278 if (alias != NULL) {
1286 nfs_set_verifier(dentry, verf);
1291 * Code common to create, mkdir, and mknod.
1293 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1294 struct nfs_fattr *fattr)
1296 struct dentry *parent = dget_parent(dentry);
1297 struct inode *dir = parent->d_inode;
1298 struct inode *inode;
1299 int error = -EACCES;
1303 /* We may have been initialized further down */
1304 if (dentry->d_inode)
1306 if (fhandle->size == 0) {
1307 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1311 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1312 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1313 struct nfs_server *server = NFS_SB(dentry->d_sb);
1314 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1318 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1319 error = PTR_ERR(inode);
1322 d_add(dentry, inode);
1327 nfs_mark_for_revalidate(dir);
1333 * Following a failed create operation, we drop the dentry rather
1334 * than retain a negative dentry. This avoids a problem in the event
1335 * that the operation succeeded on the server, but an error in the
1336 * reply path made it appear to have failed.
1338 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1339 struct nameidata *nd)
1345 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1346 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1348 attr.ia_mode = mode;
1349 attr.ia_valid = ATTR_MODE;
1351 if ((nd->flags & LOOKUP_CREATE) != 0)
1352 open_flags = nd->intent.open.flags;
1354 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1364 * See comments for nfs_proc_create regarding failed operations.
1367 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1372 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1373 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1375 if (!new_valid_dev(rdev))
1378 attr.ia_mode = mode;
1379 attr.ia_valid = ATTR_MODE;
1381 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1391 * See comments for nfs_proc_create regarding failed operations.
1393 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1398 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1399 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1401 attr.ia_valid = ATTR_MODE;
1402 attr.ia_mode = mode | S_IFDIR;
1404 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1413 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1415 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1419 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1423 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1424 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1426 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1427 /* Ensure the VFS deletes this inode */
1428 if (error == 0 && dentry->d_inode != NULL)
1429 clear_nlink(dentry->d_inode);
1430 else if (error == -ENOENT)
1431 nfs_dentry_handle_enoent(dentry);
1436 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1438 static unsigned int sillycounter;
1439 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1440 const int countersize = sizeof(sillycounter)*2;
1441 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1444 struct dentry *sdentry;
1447 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1448 dentry->d_parent->d_name.name, dentry->d_name.name,
1449 atomic_read(&dentry->d_count));
1450 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1453 * We don't allow a dentry to be silly-renamed twice.
1456 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1459 sprintf(silly, ".nfs%*.*Lx",
1460 fileidsize, fileidsize,
1461 (unsigned long long)NFS_FILEID(dentry->d_inode));
1463 /* Return delegation in anticipation of the rename */
1464 nfs_inode_return_delegation(dentry->d_inode);
1468 char *suffix = silly + slen - countersize;
1472 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1474 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1475 dentry->d_name.name, silly);
1477 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1479 * N.B. Better to return EBUSY here ... it could be
1480 * dangerous to delete the file while it's in use.
1482 if (IS_ERR(sdentry))
1484 } while(sdentry->d_inode != NULL); /* need negative lookup */
1486 qsilly.name = silly;
1487 qsilly.len = strlen(silly);
1488 if (dentry->d_inode) {
1489 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1491 nfs_mark_for_revalidate(dentry->d_inode);
1493 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1496 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1497 d_move(dentry, sdentry);
1498 error = nfs_async_unlink(dir, dentry);
1499 /* If we return 0 we don't unlink */
1507 * Remove a file after making sure there are no pending writes,
1508 * and after checking that the file has only one user.
1510 * We invalidate the attribute cache and free the inode prior to the operation
1511 * to avoid possible races if the server reuses the inode.
1513 static int nfs_safe_remove(struct dentry *dentry)
1515 struct inode *dir = dentry->d_parent->d_inode;
1516 struct inode *inode = dentry->d_inode;
1519 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1520 dentry->d_parent->d_name.name, dentry->d_name.name);
1522 /* If the dentry was sillyrenamed, we simply call d_delete() */
1523 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1528 if (inode != NULL) {
1529 nfs_inode_return_delegation(inode);
1530 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1531 /* The VFS may want to delete this inode */
1533 nfs_drop_nlink(inode);
1534 nfs_mark_for_revalidate(inode);
1536 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1537 if (error == -ENOENT)
1538 nfs_dentry_handle_enoent(dentry);
1543 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1544 * belongs to an active ".nfs..." file and we return -EBUSY.
1546 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1548 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1551 int need_rehash = 0;
1553 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1554 dir->i_ino, dentry->d_name.name);
1556 spin_lock(&dcache_lock);
1557 spin_lock(&dentry->d_lock);
1558 if (atomic_read(&dentry->d_count) > 1) {
1559 spin_unlock(&dentry->d_lock);
1560 spin_unlock(&dcache_lock);
1561 /* Start asynchronous writeout of the inode */
1562 write_inode_now(dentry->d_inode, 0);
1563 error = nfs_sillyrename(dir, dentry);
1566 if (!d_unhashed(dentry)) {
1570 spin_unlock(&dentry->d_lock);
1571 spin_unlock(&dcache_lock);
1572 error = nfs_safe_remove(dentry);
1573 if (!error || error == -ENOENT) {
1574 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1575 } else if (need_rehash)
1581 * To create a symbolic link, most file systems instantiate a new inode,
1582 * add a page to it containing the path, then write it out to the disk
1583 * using prepare_write/commit_write.
1585 * Unfortunately the NFS client can't create the in-core inode first
1586 * because it needs a file handle to create an in-core inode (see
1587 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1588 * symlink request has completed on the server.
1590 * So instead we allocate a raw page, copy the symname into it, then do
1591 * the SYMLINK request with the page as the buffer. If it succeeds, we
1592 * now have a new file handle and can instantiate an in-core NFS inode
1593 * and move the raw page into its mapping.
1595 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1597 struct pagevec lru_pvec;
1601 unsigned int pathlen = strlen(symname);
1604 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1605 dir->i_ino, dentry->d_name.name, symname);
1607 if (pathlen > PAGE_SIZE)
1608 return -ENAMETOOLONG;
1610 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1611 attr.ia_valid = ATTR_MODE;
1613 page = alloc_page(GFP_HIGHUSER);
1617 kaddr = kmap_atomic(page, KM_USER0);
1618 memcpy(kaddr, symname, pathlen);
1619 if (pathlen < PAGE_SIZE)
1620 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1621 kunmap_atomic(kaddr, KM_USER0);
1623 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1625 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1626 dir->i_sb->s_id, dir->i_ino,
1627 dentry->d_name.name, symname, error);
1634 * No big deal if we can't add this page to the page cache here.
1635 * READLINK will get the missing page from the server if needed.
1637 pagevec_init(&lru_pvec, 0);
1638 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1640 pagevec_add(&lru_pvec, page);
1641 pagevec_lru_add_file(&lru_pvec);
1642 SetPageUptodate(page);
1651 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1653 struct inode *inode = old_dentry->d_inode;
1656 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1657 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1658 dentry->d_parent->d_name.name, dentry->d_name.name);
1660 nfs_inode_return_delegation(inode);
1663 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1665 atomic_inc(&inode->i_count);
1666 d_add(dentry, inode);
1673 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1674 * different file handle for the same inode after a rename (e.g. when
1675 * moving to a different directory). A fail-safe method to do so would
1676 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1677 * rename the old file using the sillyrename stuff. This way, the original
1678 * file in old_dir will go away when the last process iput()s the inode.
1682 * It actually works quite well. One needs to have the possibility for
1683 * at least one ".nfs..." file in each directory the file ever gets
1684 * moved or linked to which happens automagically with the new
1685 * implementation that only depends on the dcache stuff instead of
1686 * using the inode layer
1688 * Unfortunately, things are a little more complicated than indicated
1689 * above. For a cross-directory move, we want to make sure we can get
1690 * rid of the old inode after the operation. This means there must be
1691 * no pending writes (if it's a file), and the use count must be 1.
1692 * If these conditions are met, we can drop the dentries before doing
1695 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1696 struct inode *new_dir, struct dentry *new_dentry)
1698 struct inode *old_inode = old_dentry->d_inode;
1699 struct inode *new_inode = new_dentry->d_inode;
1700 struct dentry *dentry = NULL, *rehash = NULL;
1703 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1704 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1705 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1706 atomic_read(&new_dentry->d_count));
1709 * For non-directories, check whether the target is busy and if so,
1710 * make a copy of the dentry and then do a silly-rename. If the
1711 * silly-rename succeeds, the copied dentry is hashed and becomes
1714 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1716 * To prevent any new references to the target during the
1717 * rename, we unhash the dentry in advance.
1719 if (!d_unhashed(new_dentry)) {
1721 rehash = new_dentry;
1724 if (atomic_read(&new_dentry->d_count) > 2) {
1727 /* copy the target dentry's name */
1728 dentry = d_alloc(new_dentry->d_parent,
1729 &new_dentry->d_name);
1733 /* silly-rename the existing target ... */
1734 err = nfs_sillyrename(new_dir, new_dentry);
1738 new_dentry = dentry;
1744 nfs_inode_return_delegation(old_inode);
1745 if (new_inode != NULL)
1746 nfs_inode_return_delegation(new_inode);
1748 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1749 new_dir, &new_dentry->d_name);
1750 nfs_mark_for_revalidate(old_inode);
1755 if (new_inode != NULL)
1756 nfs_drop_nlink(new_inode);
1757 d_move(old_dentry, new_dentry);
1758 nfs_set_verifier(new_dentry,
1759 nfs_save_change_attribute(new_dir));
1760 } else if (error == -ENOENT)
1761 nfs_dentry_handle_enoent(old_dentry);
1763 /* new dentry created? */
1769 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1770 static LIST_HEAD(nfs_access_lru_list);
1771 static atomic_long_t nfs_access_nr_entries;
1773 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1775 put_rpccred(entry->cred);
1777 smp_mb__before_atomic_dec();
1778 atomic_long_dec(&nfs_access_nr_entries);
1779 smp_mb__after_atomic_dec();
1782 static void nfs_access_free_list(struct list_head *head)
1784 struct nfs_access_entry *cache;
1786 while (!list_empty(head)) {
1787 cache = list_entry(head->next, struct nfs_access_entry, lru);
1788 list_del(&cache->lru);
1789 nfs_access_free_entry(cache);
1793 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1796 struct nfs_inode *nfsi;
1797 struct nfs_access_entry *cache;
1799 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1800 return (nr_to_scan == 0) ? 0 : -1;
1802 spin_lock(&nfs_access_lru_lock);
1803 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1804 struct inode *inode;
1806 if (nr_to_scan-- == 0)
1808 inode = &nfsi->vfs_inode;
1809 spin_lock(&inode->i_lock);
1810 if (list_empty(&nfsi->access_cache_entry_lru))
1811 goto remove_lru_entry;
1812 cache = list_entry(nfsi->access_cache_entry_lru.next,
1813 struct nfs_access_entry, lru);
1814 list_move(&cache->lru, &head);
1815 rb_erase(&cache->rb_node, &nfsi->access_cache);
1816 if (!list_empty(&nfsi->access_cache_entry_lru))
1817 list_move_tail(&nfsi->access_cache_inode_lru,
1818 &nfs_access_lru_list);
1821 list_del_init(&nfsi->access_cache_inode_lru);
1822 smp_mb__before_clear_bit();
1823 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1824 smp_mb__after_clear_bit();
1826 spin_unlock(&inode->i_lock);
1828 spin_unlock(&nfs_access_lru_lock);
1829 nfs_access_free_list(&head);
1830 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1833 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1835 struct rb_root *root_node = &nfsi->access_cache;
1837 struct nfs_access_entry *entry;
1839 /* Unhook entries from the cache */
1840 while ((n = rb_first(root_node)) != NULL) {
1841 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1842 rb_erase(n, root_node);
1843 list_move(&entry->lru, head);
1845 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1848 void nfs_access_zap_cache(struct inode *inode)
1852 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
1854 /* Remove from global LRU init */
1855 spin_lock(&nfs_access_lru_lock);
1856 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
1857 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1859 spin_lock(&inode->i_lock);
1860 __nfs_access_zap_cache(NFS_I(inode), &head);
1861 spin_unlock(&inode->i_lock);
1862 spin_unlock(&nfs_access_lru_lock);
1863 nfs_access_free_list(&head);
1866 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1868 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1869 struct nfs_access_entry *entry;
1872 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1874 if (cred < entry->cred)
1876 else if (cred > entry->cred)
1884 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1886 struct nfs_inode *nfsi = NFS_I(inode);
1887 struct nfs_access_entry *cache;
1890 spin_lock(&inode->i_lock);
1891 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1893 cache = nfs_access_search_rbtree(inode, cred);
1896 if (!nfs_have_delegated_attributes(inode) &&
1897 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1899 res->jiffies = cache->jiffies;
1900 res->cred = cache->cred;
1901 res->mask = cache->mask;
1902 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1905 spin_unlock(&inode->i_lock);
1908 rb_erase(&cache->rb_node, &nfsi->access_cache);
1909 list_del(&cache->lru);
1910 spin_unlock(&inode->i_lock);
1911 nfs_access_free_entry(cache);
1914 spin_unlock(&inode->i_lock);
1915 nfs_access_zap_cache(inode);
1919 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1921 struct nfs_inode *nfsi = NFS_I(inode);
1922 struct rb_root *root_node = &nfsi->access_cache;
1923 struct rb_node **p = &root_node->rb_node;
1924 struct rb_node *parent = NULL;
1925 struct nfs_access_entry *entry;
1927 spin_lock(&inode->i_lock);
1928 while (*p != NULL) {
1930 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1932 if (set->cred < entry->cred)
1933 p = &parent->rb_left;
1934 else if (set->cred > entry->cred)
1935 p = &parent->rb_right;
1939 rb_link_node(&set->rb_node, parent, p);
1940 rb_insert_color(&set->rb_node, root_node);
1941 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1942 spin_unlock(&inode->i_lock);
1945 rb_replace_node(parent, &set->rb_node, root_node);
1946 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1947 list_del(&entry->lru);
1948 spin_unlock(&inode->i_lock);
1949 nfs_access_free_entry(entry);
1952 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1954 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1957 RB_CLEAR_NODE(&cache->rb_node);
1958 cache->jiffies = set->jiffies;
1959 cache->cred = get_rpccred(set->cred);
1960 cache->mask = set->mask;
1962 nfs_access_add_rbtree(inode, cache);
1964 /* Update accounting */
1965 smp_mb__before_atomic_inc();
1966 atomic_long_inc(&nfs_access_nr_entries);
1967 smp_mb__after_atomic_inc();
1969 /* Add inode to global LRU list */
1970 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1971 spin_lock(&nfs_access_lru_lock);
1972 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
1973 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
1974 &nfs_access_lru_list);
1975 spin_unlock(&nfs_access_lru_lock);
1979 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1981 struct nfs_access_entry cache;
1984 status = nfs_access_get_cached(inode, cred, &cache);
1988 /* Be clever: ask server to check for all possible rights */
1989 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1991 cache.jiffies = jiffies;
1992 status = NFS_PROTO(inode)->access(inode, &cache);
1994 if (status == -ESTALE) {
1995 nfs_zap_caches(inode);
1996 if (!S_ISDIR(inode->i_mode))
1997 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2001 nfs_access_add_cache(inode, &cache);
2003 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2008 static int nfs_open_permission_mask(int openflags)
2012 if (openflags & FMODE_READ)
2014 if (openflags & FMODE_WRITE)
2016 if (openflags & FMODE_EXEC)
2021 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2023 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2026 int nfs_permission(struct inode *inode, int mask)
2028 struct rpc_cred *cred;
2031 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2033 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2035 /* Is this sys_access() ? */
2036 if (mask & (MAY_ACCESS | MAY_CHDIR))
2039 switch (inode->i_mode & S_IFMT) {
2043 /* NFSv4 has atomic_open... */
2044 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2045 && (mask & MAY_OPEN)
2046 && !(mask & MAY_EXEC))
2051 * Optimize away all write operations, since the server
2052 * will check permissions when we perform the op.
2054 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2059 if (!NFS_PROTO(inode)->access)
2062 cred = rpc_lookup_cred();
2063 if (!IS_ERR(cred)) {
2064 res = nfs_do_access(inode, cred, mask);
2067 res = PTR_ERR(cred);
2069 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2072 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2073 inode->i_sb->s_id, inode->i_ino, mask, res);
2076 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2078 res = generic_permission(inode, mask, NULL);
2084 * version-control: t
2085 * kept-new-versions: 5