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(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);
147 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
151 unsigned long page_index;
154 loff_t current_index;
155 struct nfs_entry *entry;
156 decode_dirent_t decode;
158 unsigned long timestamp;
160 } nfs_readdir_descriptor_t;
162 /* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
175 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
177 struct file *file = desc->file;
178 struct inode *inode = file->f_path.dentry->d_inode;
179 struct rpc_cred *cred = nfs_file_cred(file);
180 unsigned long timestamp;
183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __func__, (long long)desc->entry->cookie,
189 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
190 NFS_SERVER(inode)->dtsize, desc->plus);
192 /* We requested READDIRPLUS, but the server doesn't grok it */
193 if (error == -ENOTSUPP && desc->plus) {
194 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
195 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
201 desc->timestamp = timestamp;
202 desc->timestamp_valid = 1;
203 SetPageUptodate(page);
204 /* Ensure consistent page alignment of the data.
205 * Note: assumes we have exclusive access to this mapping either
206 * through inode->i_mutex or some other mechanism.
208 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
209 /* Should never happen */
210 nfs_zap_mapping(inode, inode->i_mapping);
220 int dir_decode(nfs_readdir_descriptor_t *desc)
222 __be32 *p = desc->ptr;
223 p = desc->decode(p, desc->entry, desc->plus);
227 if (desc->timestamp_valid)
228 desc->entry->fattr->time_start = desc->timestamp;
230 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
235 void dir_page_release(nfs_readdir_descriptor_t *desc)
238 page_cache_release(desc->page);
244 * Given a pointer to a buffer that has already been filled by a call
245 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
247 * If the end of the buffer has been reached, return -EAGAIN, if not,
248 * return the offset within the buffer of the next entry to be
252 int find_dirent(nfs_readdir_descriptor_t *desc)
254 struct nfs_entry *entry = desc->entry;
258 while((status = dir_decode(desc)) == 0) {
259 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
260 __func__, (unsigned long long)entry->cookie);
261 if (entry->prev_cookie == *desc->dir_cookie)
263 if (loop_count++ > 200) {
272 * Given a pointer to a buffer that has already been filled by a call
273 * to readdir, find the entry at offset 'desc->file->f_pos'.
275 * If the end of the buffer has been reached, return -EAGAIN, if not,
276 * return the offset within the buffer of the next entry to be
280 int find_dirent_index(nfs_readdir_descriptor_t *desc)
282 struct nfs_entry *entry = desc->entry;
287 status = dir_decode(desc);
291 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
292 (unsigned long long)entry->cookie, desc->current_index);
294 if (desc->file->f_pos == desc->current_index) {
295 *desc->dir_cookie = entry->cookie;
298 desc->current_index++;
299 if (loop_count++ > 200) {
308 * Find the given page, and call find_dirent() or find_dirent_index in
309 * order to try to return the next entry.
312 int find_dirent_page(nfs_readdir_descriptor_t *desc)
314 struct inode *inode = desc->file->f_path.dentry->d_inode;
318 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
319 __func__, desc->page_index,
320 (long long) *desc->dir_cookie);
322 /* If we find the page in the page_cache, we cannot be sure
323 * how fresh the data is, so we will ignore readdir_plus attributes.
325 desc->timestamp_valid = 0;
326 page = read_cache_page(inode->i_mapping, desc->page_index,
327 (filler_t *)nfs_readdir_filler, desc);
329 status = PTR_ERR(page);
333 /* NOTE: Someone else may have changed the READDIRPLUS flag */
335 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
336 if (*desc->dir_cookie != 0)
337 status = find_dirent(desc);
339 status = find_dirent_index(desc);
341 dir_page_release(desc);
343 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
348 * Recurse through the page cache pages, and return a
349 * filled nfs_entry structure of the next directory entry if possible.
351 * The target for the search is '*desc->dir_cookie' if non-0,
352 * 'desc->file->f_pos' otherwise
355 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
360 /* Always search-by-index from the beginning of the cache */
361 if (*desc->dir_cookie == 0) {
362 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
363 (long long)desc->file->f_pos);
364 desc->page_index = 0;
365 desc->entry->cookie = desc->entry->prev_cookie = 0;
366 desc->entry->eof = 0;
367 desc->current_index = 0;
369 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
370 (unsigned long long)*desc->dir_cookie);
373 res = find_dirent_page(desc);
376 /* Align to beginning of next page */
378 if (loop_count++ > 200) {
384 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
388 static inline unsigned int dt_type(struct inode *inode)
390 return (inode->i_mode >> 12) & 15;
393 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
396 * Once we've found the start of the dirent within a page: fill 'er up...
399 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
402 struct file *file = desc->file;
403 struct nfs_entry *entry = desc->entry;
404 struct dentry *dentry = NULL;
409 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
410 (unsigned long long)entry->cookie);
413 unsigned d_type = DT_UNKNOWN;
414 /* Note: entry->prev_cookie contains the cookie for
415 * retrieving the current dirent on the server */
418 /* Get a dentry if we have one */
421 dentry = nfs_readdir_lookup(desc);
423 /* Use readdirplus info */
424 if (dentry != NULL && dentry->d_inode != NULL) {
425 d_type = dt_type(dentry->d_inode);
426 fileid = NFS_FILEID(dentry->d_inode);
429 res = filldir(dirent, entry->name, entry->len,
430 file->f_pos, nfs_compat_user_ino64(fileid),
435 *desc->dir_cookie = entry->cookie;
436 if (dir_decode(desc) != 0) {
440 if (loop_count++ > 200) {
445 dir_page_release(desc);
448 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
449 (unsigned long long)*desc->dir_cookie, res);
454 * If we cannot find a cookie in our cache, we suspect that this is
455 * because it points to a deleted file, so we ask the server to return
456 * whatever it thinks is the next entry. We then feed this to filldir.
457 * If all goes well, we should then be able to find our way round the
458 * cache on the next call to readdir_search_pagecache();
460 * NOTE: we cannot add the anonymous page to the pagecache because
461 * the data it contains might not be page aligned. Besides,
462 * we should already have a complete representation of the
463 * directory in the page cache by the time we get here.
466 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
469 struct file *file = desc->file;
470 struct inode *inode = file->f_path.dentry->d_inode;
471 struct rpc_cred *cred = nfs_file_cred(file);
472 struct page *page = NULL;
474 unsigned long timestamp;
476 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
477 (unsigned long long)*desc->dir_cookie);
479 page = alloc_page(GFP_HIGHUSER);
485 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
486 *desc->dir_cookie, page,
487 NFS_SERVER(inode)->dtsize,
490 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
492 desc->timestamp = timestamp;
493 desc->timestamp_valid = 1;
494 if ((status = dir_decode(desc)) == 0)
495 desc->entry->prev_cookie = *desc->dir_cookie;
501 status = nfs_do_filldir(desc, dirent, filldir);
503 /* Reset read descriptor so it searches the page cache from
504 * the start upon the next call to readdir_search_pagecache() */
505 desc->page_index = 0;
506 desc->entry->cookie = desc->entry->prev_cookie = 0;
507 desc->entry->eof = 0;
509 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
513 dir_page_release(desc);
517 /* The file offset position represents the dirent entry number. A
518 last cookie cache takes care of the common case of reading the
521 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
523 struct dentry *dentry = filp->f_path.dentry;
524 struct inode *inode = dentry->d_inode;
525 nfs_readdir_descriptor_t my_desc,
527 struct nfs_entry my_entry;
529 struct nfs_fattr fattr;
532 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
533 dentry->d_parent->d_name.name, dentry->d_name.name,
534 (long long)filp->f_pos);
535 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
538 * filp->f_pos points to the dirent entry number.
539 * *desc->dir_cookie has the cookie for the next entry. We have
540 * to either find the entry with the appropriate number or
541 * revalidate the cookie.
543 memset(desc, 0, sizeof(*desc));
546 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
547 desc->decode = NFS_PROTO(inode)->decode_dirent;
548 desc->plus = NFS_USE_READDIRPLUS(inode);
550 my_entry.cookie = my_entry.prev_cookie = 0;
553 my_entry.fattr = &fattr;
554 nfs_fattr_init(&fattr);
555 desc->entry = &my_entry;
557 nfs_block_sillyrename(dentry);
558 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
562 while(!desc->entry->eof) {
563 res = readdir_search_pagecache(desc);
565 if (res == -EBADCOOKIE) {
566 /* This means either end of directory */
567 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
568 /* Or that the server has 'lost' a cookie */
569 res = uncached_readdir(desc, dirent, filldir);
576 if (res == -ETOOSMALL && desc->plus) {
577 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
578 nfs_zap_caches(inode);
580 desc->entry->eof = 0;
586 res = nfs_do_filldir(desc, dirent, filldir);
593 nfs_unblock_sillyrename(dentry);
596 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
597 dentry->d_parent->d_name.name, dentry->d_name.name,
602 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
604 struct dentry *dentry = filp->f_path.dentry;
605 struct inode *inode = dentry->d_inode;
607 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
608 dentry->d_parent->d_name.name,
612 mutex_lock(&inode->i_mutex);
615 offset += filp->f_pos;
623 if (offset != filp->f_pos) {
624 filp->f_pos = offset;
625 nfs_file_open_context(filp)->dir_cookie = 0;
628 mutex_unlock(&inode->i_mutex);
633 * All directory operations under NFS are synchronous, so fsync()
634 * is a dummy operation.
636 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
638 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
639 dentry->d_parent->d_name.name, dentry->d_name.name,
642 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
647 * nfs_force_lookup_revalidate - Mark the directory as having changed
648 * @dir - pointer to directory inode
650 * This forces the revalidation code in nfs_lookup_revalidate() to do a
651 * full lookup on all child dentries of 'dir' whenever a change occurs
652 * on the server that might have invalidated our dcache.
654 * The caller should be holding dir->i_lock
656 void nfs_force_lookup_revalidate(struct inode *dir)
658 NFS_I(dir)->cache_change_attribute = jiffies;
662 * A check for whether or not the parent directory has changed.
663 * In the case it has, we assume that the dentries are untrustworthy
664 * and may need to be looked up again.
666 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
670 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
672 if (!nfs_verify_change_attribute(dir, dentry->d_time))
674 /* Revalidate nfsi->cache_change_attribute before we declare a match */
675 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
677 if (!nfs_verify_change_attribute(dir, dentry->d_time))
683 * Return the intent data that applies to this particular path component
685 * Note that the current set of intents only apply to the very last
686 * component of the path.
687 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
689 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
691 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
693 return nd->flags & mask;
697 * Use intent information to check whether or not we're going to do
698 * an O_EXCL create using this path component.
700 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
702 if (NFS_PROTO(dir)->version == 2)
704 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
706 return (nd->intent.open.flags & O_EXCL) != 0;
710 * Inode and filehandle revalidation for lookups.
712 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
713 * or if the intent information indicates that we're about to open this
714 * particular file and the "nocto" mount flag is not set.
718 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
720 struct nfs_server *server = NFS_SERVER(inode);
722 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
725 /* VFS wants an on-the-wire revalidation */
726 if (nd->flags & LOOKUP_REVAL)
728 /* This is an open(2) */
729 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
730 !(server->flags & NFS_MOUNT_NOCTO) &&
731 (S_ISREG(inode->i_mode) ||
732 S_ISDIR(inode->i_mode)))
736 return nfs_revalidate_inode(server, inode);
738 return __nfs_revalidate_inode(server, inode);
742 * We judge how long we want to trust negative
743 * dentries by looking at the parent inode mtime.
745 * If parent mtime has changed, we revalidate, else we wait for a
746 * period corresponding to the parent's attribute cache timeout value.
749 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
750 struct nameidata *nd)
752 /* Don't revalidate a negative dentry if we're creating a new file */
753 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
755 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
757 return !nfs_check_verifier(dir, dentry);
761 * This is called every time the dcache has a lookup hit,
762 * and we should check whether we can really trust that
765 * NOTE! The hit can be a negative hit too, don't assume
768 * If the parent directory is seen to have changed, we throw out the
769 * cached dentry and do a new lookup.
771 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
775 struct dentry *parent;
777 struct nfs_fh fhandle;
778 struct nfs_fattr fattr;
780 parent = dget_parent(dentry);
781 dir = parent->d_inode;
782 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
783 inode = dentry->d_inode;
786 if (nfs_neg_need_reval(dir, dentry, nd))
791 if (is_bad_inode(inode)) {
792 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
793 __func__, dentry->d_parent->d_name.name,
794 dentry->d_name.name);
798 /* Force a full look up iff the parent directory has changed */
799 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
800 if (nfs_lookup_verify_inode(inode, nd))
805 if (NFS_STALE(inode))
808 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
811 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
813 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
816 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
819 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
820 __func__, dentry->d_parent->d_name.name,
821 dentry->d_name.name);
826 nfs_mark_for_revalidate(dir);
827 if (inode && S_ISDIR(inode->i_mode)) {
828 /* Purge readdir caches. */
829 nfs_zap_caches(inode);
830 /* If we have submounts, don't unhash ! */
831 if (have_submounts(dentry))
833 shrink_dcache_parent(dentry);
837 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
838 __func__, dentry->d_parent->d_name.name,
839 dentry->d_name.name);
844 * This is called from dput() when d_count is going to 0.
846 static int nfs_dentry_delete(struct dentry *dentry)
848 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
849 dentry->d_parent->d_name.name, dentry->d_name.name,
852 /* Unhash any dentry with a stale inode */
853 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
856 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
857 /* Unhash it, so that ->d_iput() would be called */
860 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
861 /* Unhash it, so that ancestors of killed async unlink
862 * files will be cleaned up during umount */
869 static void nfs_drop_nlink(struct inode *inode)
871 spin_lock(&inode->i_lock);
872 if (inode->i_nlink > 0)
874 spin_unlock(&inode->i_lock);
878 * Called when the dentry loses inode.
879 * We use it to clean up silly-renamed files.
881 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
883 if (S_ISDIR(inode->i_mode))
884 /* drop any readdir cache as it could easily be old */
885 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
887 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
889 nfs_complete_unlink(dentry, inode);
894 struct dentry_operations nfs_dentry_operations = {
895 .d_revalidate = nfs_lookup_revalidate,
896 .d_delete = nfs_dentry_delete,
897 .d_iput = nfs_dentry_iput,
900 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
903 struct dentry *parent;
904 struct inode *inode = NULL;
906 struct nfs_fh fhandle;
907 struct nfs_fattr fattr;
909 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
910 dentry->d_parent->d_name.name, dentry->d_name.name);
911 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
913 res = ERR_PTR(-ENAMETOOLONG);
914 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
917 res = ERR_PTR(-ENOMEM);
918 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
921 * If we're doing an exclusive create, optimize away the lookup
922 * but don't hash the dentry.
924 if (nfs_is_exclusive_create(dir, nd)) {
925 d_instantiate(dentry, NULL);
930 parent = dentry->d_parent;
931 /* Protect against concurrent sillydeletes */
932 nfs_block_sillyrename(parent);
933 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
934 if (error == -ENOENT)
937 res = ERR_PTR(error);
938 goto out_unblock_sillyrename;
940 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
941 res = (struct dentry *)inode;
943 goto out_unblock_sillyrename;
946 res = d_materialise_unique(dentry, inode);
949 goto out_unblock_sillyrename;
952 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
953 out_unblock_sillyrename:
954 nfs_unblock_sillyrename(parent);
960 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
962 struct dentry_operations nfs4_dentry_operations = {
963 .d_revalidate = nfs_open_revalidate,
964 .d_delete = nfs_dentry_delete,
965 .d_iput = nfs_dentry_iput,
969 * Use intent information to determine whether we need to substitute
970 * the NFSv4-style stateful OPEN for the LOOKUP call
972 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
974 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
976 /* NFS does not (yet) have a stateful open for directories */
977 if (nd->flags & LOOKUP_DIRECTORY)
979 /* Are we trying to write to a read only partition? */
980 if (__mnt_is_readonly(nd->path.mnt) &&
981 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
986 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
988 struct dentry *res = NULL;
991 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
992 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
994 /* Check that we are indeed trying to open this file */
995 if (!is_atomic_open(dir, nd))
998 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
999 res = ERR_PTR(-ENAMETOOLONG);
1002 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1004 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1006 if (nd->intent.open.flags & O_EXCL) {
1007 d_instantiate(dentry, NULL);
1011 /* Open the file on the server */
1012 res = nfs4_atomic_open(dir, dentry, nd);
1014 error = PTR_ERR(res);
1016 /* Make a negative dentry */
1020 /* This turned out not to be a regular file */
1025 if (!(nd->intent.open.flags & O_NOFOLLOW))
1031 } else if (res != NULL)
1036 return nfs_lookup(dir, dentry, nd);
1039 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1041 struct dentry *parent = NULL;
1042 struct inode *inode = dentry->d_inode;
1044 int openflags, ret = 0;
1046 parent = dget_parent(dentry);
1047 dir = parent->d_inode;
1048 if (!is_atomic_open(dir, nd))
1050 /* We can't create new files in nfs_open_revalidate(), so we
1051 * optimize away revalidation of negative dentries.
1053 if (inode == NULL) {
1054 if (!nfs_neg_need_reval(dir, dentry, nd))
1059 /* NFS only supports OPEN on regular files */
1060 if (!S_ISREG(inode->i_mode))
1062 openflags = nd->intent.open.flags;
1063 /* We cannot do exclusive creation on a positive dentry */
1064 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1066 /* We can't create new files, or truncate existing ones here */
1067 openflags &= ~(O_CREAT|O_TRUNC);
1070 * Note: we're not holding inode->i_mutex and so may be racing with
1071 * operations that change the directory. We therefore save the
1072 * change attribute *before* we do the RPC call.
1074 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1082 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1084 return nfs_lookup_revalidate(dentry, nd);
1086 #endif /* CONFIG_NFSV4 */
1088 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1090 struct dentry *parent = desc->file->f_path.dentry;
1091 struct inode *dir = parent->d_inode;
1092 struct nfs_entry *entry = desc->entry;
1093 struct dentry *dentry, *alias;
1094 struct qstr name = {
1095 .name = entry->name,
1098 struct inode *inode;
1099 unsigned long verf = nfs_save_change_attribute(dir);
1103 if (name.name[0] == '.' && name.name[1] == '.')
1104 return dget_parent(parent);
1107 if (name.name[0] == '.')
1108 return dget(parent);
1111 spin_lock(&dir->i_lock);
1112 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1113 spin_unlock(&dir->i_lock);
1116 spin_unlock(&dir->i_lock);
1118 name.hash = full_name_hash(name.name, name.len);
1119 dentry = d_lookup(parent, &name);
1120 if (dentry != NULL) {
1121 /* Is this a positive dentry that matches the readdir info? */
1122 if (dentry->d_inode != NULL &&
1123 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1124 d_mountpoint(dentry))) {
1125 if (!desc->plus || entry->fh->size == 0)
1127 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1131 /* No, so d_drop to allow one to be created */
1135 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1137 if (name.len > NFS_SERVER(dir)->namelen)
1139 /* Note: caller is already holding the dir->i_mutex! */
1140 dentry = d_alloc(parent, &name);
1143 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1144 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1145 if (IS_ERR(inode)) {
1150 alias = d_materialise_unique(dentry, inode);
1151 if (alias != NULL) {
1159 nfs_set_verifier(dentry, verf);
1164 * Code common to create, mkdir, and mknod.
1166 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1167 struct nfs_fattr *fattr)
1169 struct dentry *parent = dget_parent(dentry);
1170 struct inode *dir = parent->d_inode;
1171 struct inode *inode;
1172 int error = -EACCES;
1176 /* We may have been initialized further down */
1177 if (dentry->d_inode)
1179 if (fhandle->size == 0) {
1180 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1184 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1185 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1186 struct nfs_server *server = NFS_SB(dentry->d_sb);
1187 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1191 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1192 error = PTR_ERR(inode);
1195 d_add(dentry, inode);
1200 nfs_mark_for_revalidate(dir);
1206 * Following a failed create operation, we drop the dentry rather
1207 * than retain a negative dentry. This avoids a problem in the event
1208 * that the operation succeeded on the server, but an error in the
1209 * reply path made it appear to have failed.
1211 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1212 struct nameidata *nd)
1218 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1219 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1221 attr.ia_mode = mode;
1222 attr.ia_valid = ATTR_MODE;
1224 if ((nd->flags & LOOKUP_CREATE) != 0)
1225 open_flags = nd->intent.open.flags;
1227 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1237 * See comments for nfs_proc_create regarding failed operations.
1240 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1245 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1246 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1248 if (!new_valid_dev(rdev))
1251 attr.ia_mode = mode;
1252 attr.ia_valid = ATTR_MODE;
1254 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1264 * See comments for nfs_proc_create regarding failed operations.
1266 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1271 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1272 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1274 attr.ia_valid = ATTR_MODE;
1275 attr.ia_mode = mode | S_IFDIR;
1277 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1286 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1288 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1292 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1296 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1297 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1299 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1300 /* Ensure the VFS deletes this inode */
1301 if (error == 0 && dentry->d_inode != NULL)
1302 clear_nlink(dentry->d_inode);
1303 else if (error == -ENOENT)
1304 nfs_dentry_handle_enoent(dentry);
1309 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1311 static unsigned int sillycounter;
1312 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1313 const int countersize = sizeof(sillycounter)*2;
1314 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1317 struct dentry *sdentry;
1320 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1321 dentry->d_parent->d_name.name, dentry->d_name.name,
1322 atomic_read(&dentry->d_count));
1323 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1326 * We don't allow a dentry to be silly-renamed twice.
1329 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1332 sprintf(silly, ".nfs%*.*Lx",
1333 fileidsize, fileidsize,
1334 (unsigned long long)NFS_FILEID(dentry->d_inode));
1336 /* Return delegation in anticipation of the rename */
1337 nfs_inode_return_delegation(dentry->d_inode);
1341 char *suffix = silly + slen - countersize;
1345 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1347 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1348 dentry->d_name.name, silly);
1350 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1352 * N.B. Better to return EBUSY here ... it could be
1353 * dangerous to delete the file while it's in use.
1355 if (IS_ERR(sdentry))
1357 } while(sdentry->d_inode != NULL); /* need negative lookup */
1359 qsilly.name = silly;
1360 qsilly.len = strlen(silly);
1361 if (dentry->d_inode) {
1362 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1364 nfs_mark_for_revalidate(dentry->d_inode);
1366 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1369 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1370 d_move(dentry, sdentry);
1371 error = nfs_async_unlink(dir, dentry);
1372 /* If we return 0 we don't unlink */
1380 * Remove a file after making sure there are no pending writes,
1381 * and after checking that the file has only one user.
1383 * We invalidate the attribute cache and free the inode prior to the operation
1384 * to avoid possible races if the server reuses the inode.
1386 static int nfs_safe_remove(struct dentry *dentry)
1388 struct inode *dir = dentry->d_parent->d_inode;
1389 struct inode *inode = dentry->d_inode;
1392 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1393 dentry->d_parent->d_name.name, dentry->d_name.name);
1395 /* If the dentry was sillyrenamed, we simply call d_delete() */
1396 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1401 if (inode != NULL) {
1402 nfs_inode_return_delegation(inode);
1403 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1404 /* The VFS may want to delete this inode */
1406 nfs_drop_nlink(inode);
1407 nfs_mark_for_revalidate(inode);
1409 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1410 if (error == -ENOENT)
1411 nfs_dentry_handle_enoent(dentry);
1416 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1417 * belongs to an active ".nfs..." file and we return -EBUSY.
1419 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1421 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1424 int need_rehash = 0;
1426 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1427 dir->i_ino, dentry->d_name.name);
1429 spin_lock(&dcache_lock);
1430 spin_lock(&dentry->d_lock);
1431 if (atomic_read(&dentry->d_count) > 1) {
1432 spin_unlock(&dentry->d_lock);
1433 spin_unlock(&dcache_lock);
1434 /* Start asynchronous writeout of the inode */
1435 write_inode_now(dentry->d_inode, 0);
1436 error = nfs_sillyrename(dir, dentry);
1439 if (!d_unhashed(dentry)) {
1443 spin_unlock(&dentry->d_lock);
1444 spin_unlock(&dcache_lock);
1445 error = nfs_safe_remove(dentry);
1446 if (!error || error == -ENOENT) {
1447 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1448 } else if (need_rehash)
1454 * To create a symbolic link, most file systems instantiate a new inode,
1455 * add a page to it containing the path, then write it out to the disk
1456 * using prepare_write/commit_write.
1458 * Unfortunately the NFS client can't create the in-core inode first
1459 * because it needs a file handle to create an in-core inode (see
1460 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1461 * symlink request has completed on the server.
1463 * So instead we allocate a raw page, copy the symname into it, then do
1464 * the SYMLINK request with the page as the buffer. If it succeeds, we
1465 * now have a new file handle and can instantiate an in-core NFS inode
1466 * and move the raw page into its mapping.
1468 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1470 struct pagevec lru_pvec;
1474 unsigned int pathlen = strlen(symname);
1477 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1478 dir->i_ino, dentry->d_name.name, symname);
1480 if (pathlen > PAGE_SIZE)
1481 return -ENAMETOOLONG;
1483 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1484 attr.ia_valid = ATTR_MODE;
1486 page = alloc_page(GFP_HIGHUSER);
1490 kaddr = kmap_atomic(page, KM_USER0);
1491 memcpy(kaddr, symname, pathlen);
1492 if (pathlen < PAGE_SIZE)
1493 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1494 kunmap_atomic(kaddr, KM_USER0);
1496 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1498 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1499 dir->i_sb->s_id, dir->i_ino,
1500 dentry->d_name.name, symname, error);
1507 * No big deal if we can't add this page to the page cache here.
1508 * READLINK will get the missing page from the server if needed.
1510 pagevec_init(&lru_pvec, 0);
1511 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1513 pagevec_add(&lru_pvec, page);
1514 pagevec_lru_add(&lru_pvec);
1515 SetPageUptodate(page);
1524 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1526 struct inode *inode = old_dentry->d_inode;
1529 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1530 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1531 dentry->d_parent->d_name.name, dentry->d_name.name);
1534 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1536 atomic_inc(&inode->i_count);
1537 d_add(dentry, inode);
1544 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1545 * different file handle for the same inode after a rename (e.g. when
1546 * moving to a different directory). A fail-safe method to do so would
1547 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1548 * rename the old file using the sillyrename stuff. This way, the original
1549 * file in old_dir will go away when the last process iput()s the inode.
1553 * It actually works quite well. One needs to have the possibility for
1554 * at least one ".nfs..." file in each directory the file ever gets
1555 * moved or linked to which happens automagically with the new
1556 * implementation that only depends on the dcache stuff instead of
1557 * using the inode layer
1559 * Unfortunately, things are a little more complicated than indicated
1560 * above. For a cross-directory move, we want to make sure we can get
1561 * rid of the old inode after the operation. This means there must be
1562 * no pending writes (if it's a file), and the use count must be 1.
1563 * If these conditions are met, we can drop the dentries before doing
1566 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1567 struct inode *new_dir, struct dentry *new_dentry)
1569 struct inode *old_inode = old_dentry->d_inode;
1570 struct inode *new_inode = new_dentry->d_inode;
1571 struct dentry *dentry = NULL, *rehash = NULL;
1575 * To prevent any new references to the target during the rename,
1576 * we unhash the dentry and free the inode in advance.
1578 if (!d_unhashed(new_dentry)) {
1580 rehash = new_dentry;
1583 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1584 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1585 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1586 atomic_read(&new_dentry->d_count));
1589 * First check whether the target is busy ... we can't
1590 * safely do _any_ rename if the target is in use.
1592 * For files, make a copy of the dentry and then do a
1593 * silly-rename. If the silly-rename succeeds, the
1594 * copied dentry is hashed and becomes the new target.
1598 if (S_ISDIR(new_inode->i_mode)) {
1600 if (!S_ISDIR(old_inode->i_mode))
1602 } else if (atomic_read(&new_dentry->d_count) > 2) {
1604 /* copy the target dentry's name */
1605 dentry = d_alloc(new_dentry->d_parent,
1606 &new_dentry->d_name);
1610 /* silly-rename the existing target ... */
1611 err = nfs_sillyrename(new_dir, new_dentry);
1613 new_dentry = rehash = dentry;
1615 /* instantiate the replacement target */
1616 d_instantiate(new_dentry, NULL);
1617 } else if (atomic_read(&new_dentry->d_count) > 1)
1618 /* dentry still busy? */
1621 nfs_drop_nlink(new_inode);
1625 * ... prune child dentries and writebacks if needed.
1627 if (atomic_read(&old_dentry->d_count) > 1) {
1628 if (S_ISREG(old_inode->i_mode))
1629 nfs_wb_all(old_inode);
1630 shrink_dcache_parent(old_dentry);
1632 nfs_inode_return_delegation(old_inode);
1634 if (new_inode != NULL) {
1635 nfs_inode_return_delegation(new_inode);
1636 d_delete(new_dentry);
1639 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1640 new_dir, &new_dentry->d_name);
1641 nfs_mark_for_revalidate(old_inode);
1646 d_move(old_dentry, new_dentry);
1647 nfs_set_verifier(new_dentry,
1648 nfs_save_change_attribute(new_dir));
1649 } else if (error == -ENOENT)
1650 nfs_dentry_handle_enoent(old_dentry);
1652 /* new dentry created? */
1658 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1659 static LIST_HEAD(nfs_access_lru_list);
1660 static atomic_long_t nfs_access_nr_entries;
1662 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1664 put_rpccred(entry->cred);
1666 smp_mb__before_atomic_dec();
1667 atomic_long_dec(&nfs_access_nr_entries);
1668 smp_mb__after_atomic_dec();
1671 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1674 struct nfs_inode *nfsi;
1675 struct nfs_access_entry *cache;
1678 spin_lock(&nfs_access_lru_lock);
1679 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1680 struct rw_semaphore *s_umount;
1681 struct inode *inode;
1683 if (nr_to_scan-- == 0)
1685 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1686 if (!down_read_trylock(s_umount))
1688 inode = igrab(&nfsi->vfs_inode);
1689 if (inode == NULL) {
1693 spin_lock(&inode->i_lock);
1694 if (list_empty(&nfsi->access_cache_entry_lru))
1695 goto remove_lru_entry;
1696 cache = list_entry(nfsi->access_cache_entry_lru.next,
1697 struct nfs_access_entry, lru);
1698 list_move(&cache->lru, &head);
1699 rb_erase(&cache->rb_node, &nfsi->access_cache);
1700 if (!list_empty(&nfsi->access_cache_entry_lru))
1701 list_move_tail(&nfsi->access_cache_inode_lru,
1702 &nfs_access_lru_list);
1705 list_del_init(&nfsi->access_cache_inode_lru);
1706 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1708 spin_unlock(&inode->i_lock);
1709 spin_unlock(&nfs_access_lru_lock);
1714 spin_unlock(&nfs_access_lru_lock);
1715 while (!list_empty(&head)) {
1716 cache = list_entry(head.next, struct nfs_access_entry, lru);
1717 list_del(&cache->lru);
1718 nfs_access_free_entry(cache);
1720 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1723 static void __nfs_access_zap_cache(struct inode *inode)
1725 struct nfs_inode *nfsi = NFS_I(inode);
1726 struct rb_root *root_node = &nfsi->access_cache;
1727 struct rb_node *n, *dispose = NULL;
1728 struct nfs_access_entry *entry;
1730 /* Unhook entries from the cache */
1731 while ((n = rb_first(root_node)) != NULL) {
1732 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1733 rb_erase(n, root_node);
1734 list_del(&entry->lru);
1735 n->rb_left = dispose;
1738 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1739 spin_unlock(&inode->i_lock);
1741 /* Now kill them all! */
1742 while (dispose != NULL) {
1744 dispose = n->rb_left;
1745 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1749 void nfs_access_zap_cache(struct inode *inode)
1751 /* Remove from global LRU init */
1752 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1753 spin_lock(&nfs_access_lru_lock);
1754 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1755 spin_unlock(&nfs_access_lru_lock);
1758 spin_lock(&inode->i_lock);
1759 /* This will release the spinlock */
1760 __nfs_access_zap_cache(inode);
1763 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1765 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1766 struct nfs_access_entry *entry;
1769 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1771 if (cred < entry->cred)
1773 else if (cred > entry->cred)
1781 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1783 struct nfs_inode *nfsi = NFS_I(inode);
1784 struct nfs_access_entry *cache;
1787 spin_lock(&inode->i_lock);
1788 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1790 cache = nfs_access_search_rbtree(inode, cred);
1793 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1795 res->jiffies = cache->jiffies;
1796 res->cred = cache->cred;
1797 res->mask = cache->mask;
1798 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1801 spin_unlock(&inode->i_lock);
1804 rb_erase(&cache->rb_node, &nfsi->access_cache);
1805 list_del(&cache->lru);
1806 spin_unlock(&inode->i_lock);
1807 nfs_access_free_entry(cache);
1810 /* This will release the spinlock */
1811 __nfs_access_zap_cache(inode);
1815 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1817 struct nfs_inode *nfsi = NFS_I(inode);
1818 struct rb_root *root_node = &nfsi->access_cache;
1819 struct rb_node **p = &root_node->rb_node;
1820 struct rb_node *parent = NULL;
1821 struct nfs_access_entry *entry;
1823 spin_lock(&inode->i_lock);
1824 while (*p != NULL) {
1826 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1828 if (set->cred < entry->cred)
1829 p = &parent->rb_left;
1830 else if (set->cred > entry->cred)
1831 p = &parent->rb_right;
1835 rb_link_node(&set->rb_node, parent, p);
1836 rb_insert_color(&set->rb_node, root_node);
1837 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1838 spin_unlock(&inode->i_lock);
1841 rb_replace_node(parent, &set->rb_node, root_node);
1842 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1843 list_del(&entry->lru);
1844 spin_unlock(&inode->i_lock);
1845 nfs_access_free_entry(entry);
1848 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1850 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1853 RB_CLEAR_NODE(&cache->rb_node);
1854 cache->jiffies = set->jiffies;
1855 cache->cred = get_rpccred(set->cred);
1856 cache->mask = set->mask;
1858 nfs_access_add_rbtree(inode, cache);
1860 /* Update accounting */
1861 smp_mb__before_atomic_inc();
1862 atomic_long_inc(&nfs_access_nr_entries);
1863 smp_mb__after_atomic_inc();
1865 /* Add inode to global LRU list */
1866 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1867 spin_lock(&nfs_access_lru_lock);
1868 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1869 spin_unlock(&nfs_access_lru_lock);
1873 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1875 struct nfs_access_entry cache;
1878 status = nfs_access_get_cached(inode, cred, &cache);
1882 /* Be clever: ask server to check for all possible rights */
1883 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1885 cache.jiffies = jiffies;
1886 status = NFS_PROTO(inode)->access(inode, &cache);
1889 nfs_access_add_cache(inode, &cache);
1891 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1896 static int nfs_open_permission_mask(int openflags)
1900 if (openflags & FMODE_READ)
1902 if (openflags & FMODE_WRITE)
1904 if (openflags & FMODE_EXEC)
1909 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1911 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1914 int nfs_permission(struct inode *inode, int mask)
1916 struct rpc_cred *cred;
1919 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1921 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1923 /* Is this sys_access() ? */
1924 if (mask & MAY_ACCESS)
1927 switch (inode->i_mode & S_IFMT) {
1931 /* NFSv4 has atomic_open... */
1932 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1933 && (mask & MAY_OPEN))
1938 * Optimize away all write operations, since the server
1939 * will check permissions when we perform the op.
1941 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1946 if (!NFS_PROTO(inode)->access)
1949 cred = rpc_lookup_cred();
1950 if (!IS_ERR(cred)) {
1951 res = nfs_do_access(inode, cred, mask);
1954 res = PTR_ERR(cred);
1956 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1957 inode->i_sb->s_id, inode->i_ino, mask, res);
1960 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1962 res = generic_permission(inode, mask, NULL);
1968 * version-control: t
1969 * kept-new-versions: 5