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/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
39 #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 *, struct dentry *, 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(VFS, "NFS: opendir(%s/%ld)\n",
137 inode->i_sb->s_id, inode->i_ino);
140 /* Call generic open code in order to cache credentials */
141 res = nfs_open(inode, filp);
146 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
150 unsigned long page_index;
153 loff_t current_index;
154 struct nfs_entry *entry;
155 decode_dirent_t decode;
157 unsigned long timestamp;
159 } nfs_readdir_descriptor_t;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176 struct file *file = desc->file;
177 struct inode *inode = file->f_path.dentry->d_inode;
178 struct rpc_cred *cred = nfs_file_cred(file);
179 unsigned long timestamp;
182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __func__, (long long)desc->entry->cookie,
188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
189 NFS_SERVER(inode)->dtsize, desc->plus);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error == -ENOTSUPP && desc->plus) {
193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
200 desc->timestamp = timestamp;
201 desc->timestamp_valid = 1;
202 SetPageUptodate(page);
203 /* Ensure consistent page alignment of the data.
204 * Note: assumes we have exclusive access to this mapping either
205 * through inode->i_mutex or some other mechanism.
207 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
208 /* Should never happen */
209 nfs_zap_mapping(inode, inode->i_mapping);
219 int dir_decode(nfs_readdir_descriptor_t *desc)
221 __be32 *p = desc->ptr;
222 p = desc->decode(p, desc->entry, desc->plus);
226 if (desc->timestamp_valid)
227 desc->entry->fattr->time_start = desc->timestamp;
229 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
234 void dir_page_release(nfs_readdir_descriptor_t *desc)
237 page_cache_release(desc->page);
243 * Given a pointer to a buffer that has already been filled by a call
244 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
246 * If the end of the buffer has been reached, return -EAGAIN, if not,
247 * return the offset within the buffer of the next entry to be
251 int find_dirent(nfs_readdir_descriptor_t *desc)
253 struct nfs_entry *entry = desc->entry;
257 while((status = dir_decode(desc)) == 0) {
258 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
259 __func__, (unsigned long long)entry->cookie);
260 if (entry->prev_cookie == *desc->dir_cookie)
262 if (loop_count++ > 200) {
271 * Given a pointer to a buffer that has already been filled by a call
272 * to readdir, find the entry at offset 'desc->file->f_pos'.
274 * If the end of the buffer has been reached, return -EAGAIN, if not,
275 * return the offset within the buffer of the next entry to be
279 int find_dirent_index(nfs_readdir_descriptor_t *desc)
281 struct nfs_entry *entry = desc->entry;
286 status = dir_decode(desc);
290 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
291 (unsigned long long)entry->cookie, desc->current_index);
293 if (desc->file->f_pos == desc->current_index) {
294 *desc->dir_cookie = entry->cookie;
297 desc->current_index++;
298 if (loop_count++ > 200) {
307 * Find the given page, and call find_dirent() or find_dirent_index in
308 * order to try to return the next entry.
311 int find_dirent_page(nfs_readdir_descriptor_t *desc)
313 struct inode *inode = desc->file->f_path.dentry->d_inode;
317 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
318 __func__, desc->page_index,
319 (long long) *desc->dir_cookie);
321 /* If we find the page in the page_cache, we cannot be sure
322 * how fresh the data is, so we will ignore readdir_plus attributes.
324 desc->timestamp_valid = 0;
325 page = read_cache_page(inode->i_mapping, desc->page_index,
326 (filler_t *)nfs_readdir_filler, desc);
328 status = PTR_ERR(page);
332 /* NOTE: Someone else may have changed the READDIRPLUS flag */
334 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
335 if (*desc->dir_cookie != 0)
336 status = find_dirent(desc);
338 status = find_dirent_index(desc);
340 dir_page_release(desc);
342 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
347 * Recurse through the page cache pages, and return a
348 * filled nfs_entry structure of the next directory entry if possible.
350 * The target for the search is '*desc->dir_cookie' if non-0,
351 * 'desc->file->f_pos' otherwise
354 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
359 /* Always search-by-index from the beginning of the cache */
360 if (*desc->dir_cookie == 0) {
361 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
362 (long long)desc->file->f_pos);
363 desc->page_index = 0;
364 desc->entry->cookie = desc->entry->prev_cookie = 0;
365 desc->entry->eof = 0;
366 desc->current_index = 0;
368 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
369 (unsigned long long)*desc->dir_cookie);
372 res = find_dirent_page(desc);
375 /* Align to beginning of next page */
377 if (loop_count++ > 200) {
383 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
387 static inline unsigned int dt_type(struct inode *inode)
389 return (inode->i_mode >> 12) & 15;
392 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
395 * Once we've found the start of the dirent within a page: fill 'er up...
398 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
401 struct file *file = desc->file;
402 struct nfs_entry *entry = desc->entry;
403 struct dentry *dentry = NULL;
408 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
409 (unsigned long long)entry->cookie);
412 unsigned d_type = DT_UNKNOWN;
413 /* Note: entry->prev_cookie contains the cookie for
414 * retrieving the current dirent on the server */
417 /* Get a dentry if we have one */
420 dentry = nfs_readdir_lookup(desc);
422 /* Use readdirplus info */
423 if (dentry != NULL && dentry->d_inode != NULL) {
424 d_type = dt_type(dentry->d_inode);
425 fileid = NFS_FILEID(dentry->d_inode);
428 res = filldir(dirent, entry->name, entry->len,
429 file->f_pos, nfs_compat_user_ino64(fileid),
434 *desc->dir_cookie = entry->cookie;
435 if (dir_decode(desc) != 0) {
439 if (loop_count++ > 200) {
444 dir_page_release(desc);
447 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
448 (unsigned long long)*desc->dir_cookie, res);
453 * If we cannot find a cookie in our cache, we suspect that this is
454 * because it points to a deleted file, so we ask the server to return
455 * whatever it thinks is the next entry. We then feed this to filldir.
456 * If all goes well, we should then be able to find our way round the
457 * cache on the next call to readdir_search_pagecache();
459 * NOTE: we cannot add the anonymous page to the pagecache because
460 * the data it contains might not be page aligned. Besides,
461 * we should already have a complete representation of the
462 * directory in the page cache by the time we get here.
465 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
468 struct file *file = desc->file;
469 struct inode *inode = file->f_path.dentry->d_inode;
470 struct rpc_cred *cred = nfs_file_cred(file);
471 struct page *page = NULL;
473 unsigned long timestamp;
475 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
476 (unsigned long long)*desc->dir_cookie);
478 page = alloc_page(GFP_HIGHUSER);
484 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
485 *desc->dir_cookie, page,
486 NFS_SERVER(inode)->dtsize,
489 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
491 desc->timestamp = timestamp;
492 desc->timestamp_valid = 1;
493 if ((status = dir_decode(desc)) == 0)
494 desc->entry->prev_cookie = *desc->dir_cookie;
500 status = nfs_do_filldir(desc, dirent, filldir);
502 /* Reset read descriptor so it searches the page cache from
503 * the start upon the next call to readdir_search_pagecache() */
504 desc->page_index = 0;
505 desc->entry->cookie = desc->entry->prev_cookie = 0;
506 desc->entry->eof = 0;
508 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
512 dir_page_release(desc);
516 /* The file offset position represents the dirent entry number. A
517 last cookie cache takes care of the common case of reading the
520 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
522 struct dentry *dentry = filp->f_path.dentry;
523 struct inode *inode = dentry->d_inode;
524 nfs_readdir_descriptor_t my_desc,
526 struct nfs_entry my_entry;
528 struct nfs_fattr fattr;
531 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
532 dentry->d_parent->d_name.name, dentry->d_name.name,
533 (long long)filp->f_pos);
534 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
539 * filp->f_pos points to the dirent entry number.
540 * *desc->dir_cookie has the cookie for the next entry. We have
541 * to either find the entry with the appropriate number or
542 * revalidate the cookie.
544 memset(desc, 0, sizeof(*desc));
547 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
548 desc->decode = NFS_PROTO(inode)->decode_dirent;
549 desc->plus = NFS_USE_READDIRPLUS(inode);
551 my_entry.cookie = my_entry.prev_cookie = 0;
554 my_entry.fattr = &fattr;
555 nfs_fattr_init(&fattr);
556 desc->entry = &my_entry;
558 nfs_block_sillyrename(dentry);
559 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
563 while(!desc->entry->eof) {
564 res = readdir_search_pagecache(desc);
566 if (res == -EBADCOOKIE) {
567 /* This means either end of directory */
568 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
569 /* Or that the server has 'lost' a cookie */
570 res = uncached_readdir(desc, dirent, filldir);
577 if (res == -ETOOSMALL && desc->plus) {
578 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
579 nfs_zap_caches(inode);
581 desc->entry->eof = 0;
587 res = nfs_do_filldir(desc, dirent, filldir);
594 nfs_unblock_sillyrename(dentry);
598 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
599 dentry->d_parent->d_name.name, dentry->d_name.name,
604 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
606 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
609 offset += filp->f_pos;
617 if (offset != filp->f_pos) {
618 filp->f_pos = offset;
619 nfs_file_open_context(filp)->dir_cookie = 0;
622 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
627 * All directory operations under NFS are synchronous, so fsync()
628 * is a dummy operation.
630 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
632 dfprintk(VFS, "NFS: fsync dir(%s/%s) datasync %d\n",
633 dentry->d_parent->d_name.name, dentry->d_name.name,
636 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
641 * nfs_force_lookup_revalidate - Mark the directory as having changed
642 * @dir - pointer to directory inode
644 * This forces the revalidation code in nfs_lookup_revalidate() to do a
645 * full lookup on all child dentries of 'dir' whenever a change occurs
646 * on the server that might have invalidated our dcache.
648 * The caller should be holding dir->i_lock
650 void nfs_force_lookup_revalidate(struct inode *dir)
652 NFS_I(dir)->cache_change_attribute = jiffies;
656 * A check for whether or not the parent directory has changed.
657 * In the case it has, we assume that the dentries are untrustworthy
658 * and may need to be looked up again.
660 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
664 if (!nfs_verify_change_attribute(dir, dentry->d_time))
666 /* Revalidate nfsi->cache_change_attribute before we declare a match */
667 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
669 if (!nfs_verify_change_attribute(dir, dentry->d_time))
675 * Return the intent data that applies to this particular path component
677 * Note that the current set of intents only apply to the very last
678 * component of the path.
679 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
681 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
683 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
685 return nd->flags & mask;
689 * Use intent information to check whether or not we're going to do
690 * an O_EXCL create using this path component.
692 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
694 if (NFS_PROTO(dir)->version == 2)
696 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
698 return (nd->intent.open.flags & O_EXCL) != 0;
702 * Inode and filehandle revalidation for lookups.
704 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
705 * or if the intent information indicates that we're about to open this
706 * particular file and the "nocto" mount flag is not set.
710 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
712 struct nfs_server *server = NFS_SERVER(inode);
714 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
717 /* VFS wants an on-the-wire revalidation */
718 if (nd->flags & LOOKUP_REVAL)
720 /* This is an open(2) */
721 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
722 !(server->flags & NFS_MOUNT_NOCTO) &&
723 (S_ISREG(inode->i_mode) ||
724 S_ISDIR(inode->i_mode)))
728 return nfs_revalidate_inode(server, inode);
730 return __nfs_revalidate_inode(server, inode);
734 * We judge how long we want to trust negative
735 * dentries by looking at the parent inode mtime.
737 * If parent mtime has changed, we revalidate, else we wait for a
738 * period corresponding to the parent's attribute cache timeout value.
741 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
742 struct nameidata *nd)
744 /* Don't revalidate a negative dentry if we're creating a new file */
745 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
747 return !nfs_check_verifier(dir, dentry);
751 * This is called every time the dcache has a lookup hit,
752 * and we should check whether we can really trust that
755 * NOTE! The hit can be a negative hit too, don't assume
758 * If the parent directory is seen to have changed, we throw out the
759 * cached dentry and do a new lookup.
761 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
765 struct dentry *parent;
767 struct nfs_fh fhandle;
768 struct nfs_fattr fattr;
770 parent = dget_parent(dentry);
772 dir = parent->d_inode;
773 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
774 inode = dentry->d_inode;
777 if (nfs_neg_need_reval(dir, dentry, nd))
782 if (is_bad_inode(inode)) {
783 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
784 __func__, dentry->d_parent->d_name.name,
785 dentry->d_name.name);
789 /* Force a full look up iff the parent directory has changed */
790 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
791 if (nfs_lookup_verify_inode(inode, nd))
796 if (NFS_STALE(inode))
799 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
802 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
804 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
807 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
812 __func__, dentry->d_parent->d_name.name,
813 dentry->d_name.name);
818 nfs_mark_for_revalidate(dir);
819 if (inode && S_ISDIR(inode->i_mode)) {
820 /* Purge readdir caches. */
821 nfs_zap_caches(inode);
822 /* If we have submounts, don't unhash ! */
823 if (have_submounts(dentry))
825 shrink_dcache_parent(dentry);
830 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
831 __func__, dentry->d_parent->d_name.name,
832 dentry->d_name.name);
837 * This is called from dput() when d_count is going to 0.
839 static int nfs_dentry_delete(struct dentry *dentry)
841 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
842 dentry->d_parent->d_name.name, dentry->d_name.name,
845 /* Unhash any dentry with a stale inode */
846 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
849 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
850 /* Unhash it, so that ->d_iput() would be called */
853 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
854 /* Unhash it, so that ancestors of killed async unlink
855 * files will be cleaned up during umount */
863 * Called when the dentry loses inode.
864 * We use it to clean up silly-renamed files.
866 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
868 if (S_ISDIR(inode->i_mode))
869 /* drop any readdir cache as it could easily be old */
870 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
872 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
875 nfs_complete_unlink(dentry, inode);
881 struct dentry_operations nfs_dentry_operations = {
882 .d_revalidate = nfs_lookup_revalidate,
883 .d_delete = nfs_dentry_delete,
884 .d_iput = nfs_dentry_iput,
887 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
890 struct dentry *parent;
891 struct inode *inode = NULL;
893 struct nfs_fh fhandle;
894 struct nfs_fattr fattr;
896 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
897 dentry->d_parent->d_name.name, dentry->d_name.name);
898 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
900 res = ERR_PTR(-ENAMETOOLONG);
901 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
904 res = ERR_PTR(-ENOMEM);
905 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
910 * If we're doing an exclusive create, optimize away the lookup
911 * but don't hash the dentry.
913 if (nfs_is_exclusive_create(dir, nd)) {
914 d_instantiate(dentry, NULL);
919 parent = dentry->d_parent;
920 /* Protect against concurrent sillydeletes */
921 nfs_block_sillyrename(parent);
922 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
923 if (error == -ENOENT)
926 res = ERR_PTR(error);
927 goto out_unblock_sillyrename;
929 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
930 res = (struct dentry *)inode;
932 goto out_unblock_sillyrename;
935 res = d_materialise_unique(dentry, inode);
938 goto out_unblock_sillyrename;
941 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
942 out_unblock_sillyrename:
943 nfs_unblock_sillyrename(parent);
951 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
953 struct dentry_operations nfs4_dentry_operations = {
954 .d_revalidate = nfs_open_revalidate,
955 .d_delete = nfs_dentry_delete,
956 .d_iput = nfs_dentry_iput,
960 * Use intent information to determine whether we need to substitute
961 * the NFSv4-style stateful OPEN for the LOOKUP call
963 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
965 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
967 /* NFS does not (yet) have a stateful open for directories */
968 if (nd->flags & LOOKUP_DIRECTORY)
970 /* Are we trying to write to a read only partition? */
971 if (__mnt_is_readonly(nd->path.mnt) &&
972 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
977 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
979 struct dentry *res = NULL;
982 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
983 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
985 /* Check that we are indeed trying to open this file */
986 if (!is_atomic_open(dir, nd))
989 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
990 res = ERR_PTR(-ENAMETOOLONG);
993 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
995 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
997 if (nd->intent.open.flags & O_EXCL) {
998 d_instantiate(dentry, NULL);
1002 /* Open the file on the server */
1004 res = nfs4_atomic_open(dir, dentry, nd);
1007 error = PTR_ERR(res);
1009 /* Make a negative dentry */
1013 /* This turned out not to be a regular file */
1018 if (!(nd->intent.open.flags & O_NOFOLLOW))
1024 } else if (res != NULL)
1029 return nfs_lookup(dir, dentry, nd);
1032 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1034 struct dentry *parent = NULL;
1035 struct inode *inode = dentry->d_inode;
1037 int openflags, ret = 0;
1039 parent = dget_parent(dentry);
1040 dir = parent->d_inode;
1041 if (!is_atomic_open(dir, nd))
1043 /* We can't create new files in nfs_open_revalidate(), so we
1044 * optimize away revalidation of negative dentries.
1046 if (inode == NULL) {
1047 if (!nfs_neg_need_reval(dir, dentry, nd))
1052 /* NFS only supports OPEN on regular files */
1053 if (!S_ISREG(inode->i_mode))
1055 openflags = nd->intent.open.flags;
1056 /* We cannot do exclusive creation on a positive dentry */
1057 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1059 /* We can't create new files, or truncate existing ones here */
1060 openflags &= ~(O_CREAT|O_TRUNC);
1063 * Note: we're not holding inode->i_mutex and so may be racing with
1064 * operations that change the directory. We therefore save the
1065 * change attribute *before* we do the RPC call.
1068 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1077 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1079 return nfs_lookup_revalidate(dentry, nd);
1081 #endif /* CONFIG_NFSV4 */
1083 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1085 struct dentry *parent = desc->file->f_path.dentry;
1086 struct inode *dir = parent->d_inode;
1087 struct nfs_entry *entry = desc->entry;
1088 struct dentry *dentry, *alias;
1089 struct qstr name = {
1090 .name = entry->name,
1093 struct inode *inode;
1094 unsigned long verf = nfs_save_change_attribute(dir);
1098 if (name.name[0] == '.' && name.name[1] == '.')
1099 return dget_parent(parent);
1102 if (name.name[0] == '.')
1103 return dget(parent);
1106 spin_lock(&dir->i_lock);
1107 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1108 spin_unlock(&dir->i_lock);
1111 spin_unlock(&dir->i_lock);
1113 name.hash = full_name_hash(name.name, name.len);
1114 dentry = d_lookup(parent, &name);
1115 if (dentry != NULL) {
1116 /* Is this a positive dentry that matches the readdir info? */
1117 if (dentry->d_inode != NULL &&
1118 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1119 d_mountpoint(dentry))) {
1120 if (!desc->plus || entry->fh->size == 0)
1122 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1126 /* No, so d_drop to allow one to be created */
1130 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1132 if (name.len > NFS_SERVER(dir)->namelen)
1134 /* Note: caller is already holding the dir->i_mutex! */
1135 dentry = d_alloc(parent, &name);
1138 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1139 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1140 if (IS_ERR(inode)) {
1145 alias = d_materialise_unique(dentry, inode);
1146 if (alias != NULL) {
1154 nfs_set_verifier(dentry, verf);
1159 * Code common to create, mkdir, and mknod.
1161 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1162 struct nfs_fattr *fattr)
1164 struct dentry *parent = dget_parent(dentry);
1165 struct inode *dir = parent->d_inode;
1166 struct inode *inode;
1167 int error = -EACCES;
1171 /* We may have been initialized further down */
1172 if (dentry->d_inode)
1174 if (fhandle->size == 0) {
1175 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1179 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1180 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1181 struct nfs_server *server = NFS_SB(dentry->d_sb);
1182 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1186 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1187 error = PTR_ERR(inode);
1190 d_add(dentry, inode);
1195 nfs_mark_for_revalidate(dir);
1201 * Following a failed create operation, we drop the dentry rather
1202 * than retain a negative dentry. This avoids a problem in the event
1203 * that the operation succeeded on the server, but an error in the
1204 * reply path made it appear to have failed.
1206 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1207 struct nameidata *nd)
1213 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1214 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1216 attr.ia_mode = mode;
1217 attr.ia_valid = ATTR_MODE;
1219 if ((nd->flags & LOOKUP_CREATE) != 0)
1220 open_flags = nd->intent.open.flags;
1223 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1235 * See comments for nfs_proc_create regarding failed operations.
1238 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1243 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1244 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1246 if (!new_valid_dev(rdev))
1249 attr.ia_mode = mode;
1250 attr.ia_valid = ATTR_MODE;
1253 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1265 * See comments for nfs_proc_create regarding failed operations.
1267 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1272 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1273 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1275 attr.ia_valid = ATTR_MODE;
1276 attr.ia_mode = mode | S_IFDIR;
1279 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1290 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1292 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1296 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1300 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1301 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1304 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1305 /* Ensure the VFS deletes this inode */
1306 if (error == 0 && dentry->d_inode != NULL)
1307 clear_nlink(dentry->d_inode);
1308 else if (error == -ENOENT)
1309 nfs_dentry_handle_enoent(dentry);
1315 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1317 static unsigned int sillycounter;
1318 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1319 const int countersize = sizeof(sillycounter)*2;
1320 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1323 struct dentry *sdentry;
1326 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1327 dentry->d_parent->d_name.name, dentry->d_name.name,
1328 atomic_read(&dentry->d_count));
1329 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1332 * We don't allow a dentry to be silly-renamed twice.
1335 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1338 sprintf(silly, ".nfs%*.*Lx",
1339 fileidsize, fileidsize,
1340 (unsigned long long)NFS_FILEID(dentry->d_inode));
1342 /* Return delegation in anticipation of the rename */
1343 nfs_inode_return_delegation(dentry->d_inode);
1347 char *suffix = silly + slen - countersize;
1351 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1353 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1354 dentry->d_name.name, silly);
1356 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1358 * N.B. Better to return EBUSY here ... it could be
1359 * dangerous to delete the file while it's in use.
1361 if (IS_ERR(sdentry))
1363 } while(sdentry->d_inode != NULL); /* need negative lookup */
1365 qsilly.name = silly;
1366 qsilly.len = strlen(silly);
1367 if (dentry->d_inode) {
1368 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1370 nfs_mark_for_revalidate(dentry->d_inode);
1372 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1375 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1376 d_move(dentry, sdentry);
1377 error = nfs_async_unlink(dir, dentry);
1378 /* If we return 0 we don't unlink */
1386 * Remove a file after making sure there are no pending writes,
1387 * and after checking that the file has only one user.
1389 * We invalidate the attribute cache and free the inode prior to the operation
1390 * to avoid possible races if the server reuses the inode.
1392 static int nfs_safe_remove(struct dentry *dentry)
1394 struct inode *dir = dentry->d_parent->d_inode;
1395 struct inode *inode = dentry->d_inode;
1398 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1399 dentry->d_parent->d_name.name, dentry->d_name.name);
1401 /* If the dentry was sillyrenamed, we simply call d_delete() */
1402 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1407 if (inode != NULL) {
1408 nfs_inode_return_delegation(inode);
1409 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1410 /* The VFS may want to delete this inode */
1413 nfs_mark_for_revalidate(inode);
1415 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1416 if (error == -ENOENT)
1417 nfs_dentry_handle_enoent(dentry);
1422 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1423 * belongs to an active ".nfs..." file and we return -EBUSY.
1425 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1427 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1430 int need_rehash = 0;
1432 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1433 dir->i_ino, dentry->d_name.name);
1436 spin_lock(&dcache_lock);
1437 spin_lock(&dentry->d_lock);
1438 if (atomic_read(&dentry->d_count) > 1) {
1439 spin_unlock(&dentry->d_lock);
1440 spin_unlock(&dcache_lock);
1441 /* Start asynchronous writeout of the inode */
1442 write_inode_now(dentry->d_inode, 0);
1443 error = nfs_sillyrename(dir, dentry);
1447 if (!d_unhashed(dentry)) {
1451 spin_unlock(&dentry->d_lock);
1452 spin_unlock(&dcache_lock);
1453 error = nfs_safe_remove(dentry);
1454 if (!error || error == -ENOENT) {
1455 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1456 } else if (need_rehash)
1463 * To create a symbolic link, most file systems instantiate a new inode,
1464 * add a page to it containing the path, then write it out to the disk
1465 * using prepare_write/commit_write.
1467 * Unfortunately the NFS client can't create the in-core inode first
1468 * because it needs a file handle to create an in-core inode (see
1469 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1470 * symlink request has completed on the server.
1472 * So instead we allocate a raw page, copy the symname into it, then do
1473 * the SYMLINK request with the page as the buffer. If it succeeds, we
1474 * now have a new file handle and can instantiate an in-core NFS inode
1475 * and move the raw page into its mapping.
1477 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1479 struct pagevec lru_pvec;
1483 unsigned int pathlen = strlen(symname);
1486 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1487 dir->i_ino, dentry->d_name.name, symname);
1489 if (pathlen > PAGE_SIZE)
1490 return -ENAMETOOLONG;
1492 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1493 attr.ia_valid = ATTR_MODE;
1497 page = alloc_page(GFP_HIGHUSER);
1503 kaddr = kmap_atomic(page, KM_USER0);
1504 memcpy(kaddr, symname, pathlen);
1505 if (pathlen < PAGE_SIZE)
1506 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1507 kunmap_atomic(kaddr, KM_USER0);
1509 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1511 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1512 dir->i_sb->s_id, dir->i_ino,
1513 dentry->d_name.name, symname, error);
1521 * No big deal if we can't add this page to the page cache here.
1522 * READLINK will get the missing page from the server if needed.
1524 pagevec_init(&lru_pvec, 0);
1525 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1527 pagevec_add(&lru_pvec, page);
1528 pagevec_lru_add(&lru_pvec);
1529 SetPageUptodate(page);
1539 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1541 struct inode *inode = old_dentry->d_inode;
1544 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1545 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1546 dentry->d_parent->d_name.name, dentry->d_name.name);
1550 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1552 atomic_inc(&inode->i_count);
1553 d_add(dentry, inode);
1561 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1562 * different file handle for the same inode after a rename (e.g. when
1563 * moving to a different directory). A fail-safe method to do so would
1564 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1565 * rename the old file using the sillyrename stuff. This way, the original
1566 * file in old_dir will go away when the last process iput()s the inode.
1570 * It actually works quite well. One needs to have the possibility for
1571 * at least one ".nfs..." file in each directory the file ever gets
1572 * moved or linked to which happens automagically with the new
1573 * implementation that only depends on the dcache stuff instead of
1574 * using the inode layer
1576 * Unfortunately, things are a little more complicated than indicated
1577 * above. For a cross-directory move, we want to make sure we can get
1578 * rid of the old inode after the operation. This means there must be
1579 * no pending writes (if it's a file), and the use count must be 1.
1580 * If these conditions are met, we can drop the dentries before doing
1583 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1584 struct inode *new_dir, struct dentry *new_dentry)
1586 struct inode *old_inode = old_dentry->d_inode;
1587 struct inode *new_inode = new_dentry->d_inode;
1588 struct dentry *dentry = NULL, *rehash = NULL;
1592 * To prevent any new references to the target during the rename,
1593 * we unhash the dentry and free the inode in advance.
1596 if (!d_unhashed(new_dentry)) {
1598 rehash = new_dentry;
1601 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1602 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1603 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1604 atomic_read(&new_dentry->d_count));
1607 * First check whether the target is busy ... we can't
1608 * safely do _any_ rename if the target is in use.
1610 * For files, make a copy of the dentry and then do a
1611 * silly-rename. If the silly-rename succeeds, the
1612 * copied dentry is hashed and becomes the new target.
1616 if (S_ISDIR(new_inode->i_mode)) {
1618 if (!S_ISDIR(old_inode->i_mode))
1620 } else if (atomic_read(&new_dentry->d_count) > 2) {
1622 /* copy the target dentry's name */
1623 dentry = d_alloc(new_dentry->d_parent,
1624 &new_dentry->d_name);
1628 /* silly-rename the existing target ... */
1629 err = nfs_sillyrename(new_dir, new_dentry);
1631 new_dentry = rehash = dentry;
1633 /* instantiate the replacement target */
1634 d_instantiate(new_dentry, NULL);
1635 } else if (atomic_read(&new_dentry->d_count) > 1)
1636 /* dentry still busy? */
1639 drop_nlink(new_inode);
1643 * ... prune child dentries and writebacks if needed.
1645 if (atomic_read(&old_dentry->d_count) > 1) {
1646 if (S_ISREG(old_inode->i_mode))
1647 nfs_wb_all(old_inode);
1648 shrink_dcache_parent(old_dentry);
1650 nfs_inode_return_delegation(old_inode);
1652 if (new_inode != NULL) {
1653 nfs_inode_return_delegation(new_inode);
1654 d_delete(new_dentry);
1657 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1658 new_dir, &new_dentry->d_name);
1659 nfs_mark_for_revalidate(old_inode);
1664 d_move(old_dentry, new_dentry);
1665 nfs_set_verifier(new_dentry,
1666 nfs_save_change_attribute(new_dir));
1667 } else if (error == -ENOENT)
1668 nfs_dentry_handle_enoent(old_dentry);
1670 /* new dentry created? */
1677 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1678 static LIST_HEAD(nfs_access_lru_list);
1679 static atomic_long_t nfs_access_nr_entries;
1681 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1683 put_rpccred(entry->cred);
1685 smp_mb__before_atomic_dec();
1686 atomic_long_dec(&nfs_access_nr_entries);
1687 smp_mb__after_atomic_dec();
1690 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1693 struct nfs_inode *nfsi;
1694 struct nfs_access_entry *cache;
1697 spin_lock(&nfs_access_lru_lock);
1698 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1699 struct rw_semaphore *s_umount;
1700 struct inode *inode;
1702 if (nr_to_scan-- == 0)
1704 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1705 if (!down_read_trylock(s_umount))
1707 inode = igrab(&nfsi->vfs_inode);
1708 if (inode == NULL) {
1712 spin_lock(&inode->i_lock);
1713 if (list_empty(&nfsi->access_cache_entry_lru))
1714 goto remove_lru_entry;
1715 cache = list_entry(nfsi->access_cache_entry_lru.next,
1716 struct nfs_access_entry, lru);
1717 list_move(&cache->lru, &head);
1718 rb_erase(&cache->rb_node, &nfsi->access_cache);
1719 if (!list_empty(&nfsi->access_cache_entry_lru))
1720 list_move_tail(&nfsi->access_cache_inode_lru,
1721 &nfs_access_lru_list);
1724 list_del_init(&nfsi->access_cache_inode_lru);
1725 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1727 spin_unlock(&inode->i_lock);
1728 spin_unlock(&nfs_access_lru_lock);
1733 spin_unlock(&nfs_access_lru_lock);
1734 while (!list_empty(&head)) {
1735 cache = list_entry(head.next, struct nfs_access_entry, lru);
1736 list_del(&cache->lru);
1737 nfs_access_free_entry(cache);
1739 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1742 static void __nfs_access_zap_cache(struct inode *inode)
1744 struct nfs_inode *nfsi = NFS_I(inode);
1745 struct rb_root *root_node = &nfsi->access_cache;
1746 struct rb_node *n, *dispose = NULL;
1747 struct nfs_access_entry *entry;
1749 /* Unhook entries from the cache */
1750 while ((n = rb_first(root_node)) != NULL) {
1751 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1752 rb_erase(n, root_node);
1753 list_del(&entry->lru);
1754 n->rb_left = dispose;
1757 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1758 spin_unlock(&inode->i_lock);
1760 /* Now kill them all! */
1761 while (dispose != NULL) {
1763 dispose = n->rb_left;
1764 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1768 void nfs_access_zap_cache(struct inode *inode)
1770 /* Remove from global LRU init */
1771 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1772 spin_lock(&nfs_access_lru_lock);
1773 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1774 spin_unlock(&nfs_access_lru_lock);
1777 spin_lock(&inode->i_lock);
1778 /* This will release the spinlock */
1779 __nfs_access_zap_cache(inode);
1782 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1784 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1785 struct nfs_access_entry *entry;
1788 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1790 if (cred < entry->cred)
1792 else if (cred > entry->cred)
1800 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1802 struct nfs_inode *nfsi = NFS_I(inode);
1803 struct nfs_access_entry *cache;
1806 spin_lock(&inode->i_lock);
1807 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1809 cache = nfs_access_search_rbtree(inode, cred);
1812 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1814 res->jiffies = cache->jiffies;
1815 res->cred = cache->cred;
1816 res->mask = cache->mask;
1817 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1820 spin_unlock(&inode->i_lock);
1823 rb_erase(&cache->rb_node, &nfsi->access_cache);
1824 list_del(&cache->lru);
1825 spin_unlock(&inode->i_lock);
1826 nfs_access_free_entry(cache);
1829 /* This will release the spinlock */
1830 __nfs_access_zap_cache(inode);
1834 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1836 struct nfs_inode *nfsi = NFS_I(inode);
1837 struct rb_root *root_node = &nfsi->access_cache;
1838 struct rb_node **p = &root_node->rb_node;
1839 struct rb_node *parent = NULL;
1840 struct nfs_access_entry *entry;
1842 spin_lock(&inode->i_lock);
1843 while (*p != NULL) {
1845 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1847 if (set->cred < entry->cred)
1848 p = &parent->rb_left;
1849 else if (set->cred > entry->cred)
1850 p = &parent->rb_right;
1854 rb_link_node(&set->rb_node, parent, p);
1855 rb_insert_color(&set->rb_node, root_node);
1856 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1857 spin_unlock(&inode->i_lock);
1860 rb_replace_node(parent, &set->rb_node, root_node);
1861 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1862 list_del(&entry->lru);
1863 spin_unlock(&inode->i_lock);
1864 nfs_access_free_entry(entry);
1867 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1869 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1872 RB_CLEAR_NODE(&cache->rb_node);
1873 cache->jiffies = set->jiffies;
1874 cache->cred = get_rpccred(set->cred);
1875 cache->mask = set->mask;
1877 nfs_access_add_rbtree(inode, cache);
1879 /* Update accounting */
1880 smp_mb__before_atomic_inc();
1881 atomic_long_inc(&nfs_access_nr_entries);
1882 smp_mb__after_atomic_inc();
1884 /* Add inode to global LRU list */
1885 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1886 spin_lock(&nfs_access_lru_lock);
1887 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1888 spin_unlock(&nfs_access_lru_lock);
1892 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1894 struct nfs_access_entry cache;
1897 status = nfs_access_get_cached(inode, cred, &cache);
1901 /* Be clever: ask server to check for all possible rights */
1902 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1904 cache.jiffies = jiffies;
1905 status = NFS_PROTO(inode)->access(inode, &cache);
1908 nfs_access_add_cache(inode, &cache);
1910 if ((cache.mask & mask) == mask)
1915 static int nfs_open_permission_mask(int openflags)
1919 if (openflags & FMODE_READ)
1921 if (openflags & FMODE_WRITE)
1923 if (openflags & FMODE_EXEC)
1928 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1930 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1933 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1935 struct rpc_cred *cred;
1938 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1942 /* Is this sys_access() ? */
1943 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1946 switch (inode->i_mode & S_IFMT) {
1950 /* NFSv4 has atomic_open... */
1951 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1953 && (nd->flags & LOOKUP_OPEN))
1958 * Optimize away all write operations, since the server
1959 * will check permissions when we perform the op.
1961 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1968 if (!NFS_PROTO(inode)->access)
1971 cred = rpc_lookup_cred();
1972 if (!IS_ERR(cred)) {
1973 res = nfs_do_access(inode, cred, mask);
1976 res = PTR_ERR(cred);
1979 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1980 inode->i_sb->s_id, inode->i_ino, mask, res);
1983 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1985 res = generic_permission(inode, mask, NULL);
1992 * version-control: t
1993 * kept-new-versions: 5