4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
38 #include "delegation.h"
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode *, struct file *);
45 static int nfs_readdir(struct file *, void *, filldir_t);
46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
48 static int nfs_mkdir(struct inode *, struct dentry *, int);
49 static int nfs_rmdir(struct inode *, struct dentry *);
50 static int nfs_unlink(struct inode *, struct dentry *);
51 static int nfs_symlink(struct inode *, struct dentry *, const char *);
52 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
54 static int nfs_rename(struct inode *, struct dentry *,
55 struct inode *, struct dentry *);
56 static int nfs_fsync_dir(struct file *, struct dentry *, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 const struct file_operations nfs_dir_operations = {
60 .llseek = nfs_llseek_dir,
61 .read = generic_read_dir,
62 .readdir = nfs_readdir,
64 .release = nfs_release,
65 .fsync = nfs_fsync_dir,
68 const struct inode_operations nfs_dir_inode_operations = {
73 .symlink = nfs_symlink,
78 .permission = nfs_permission,
79 .getattr = nfs_getattr,
80 .setattr = nfs_setattr,
84 const struct inode_operations nfs3_dir_inode_operations = {
89 .symlink = nfs_symlink,
94 .permission = nfs_permission,
95 .getattr = nfs_getattr,
96 .setattr = nfs_setattr,
97 .listxattr = nfs3_listxattr,
98 .getxattr = nfs3_getxattr,
99 .setxattr = nfs3_setxattr,
100 .removexattr = nfs3_removexattr,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107 const struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode *inode, struct file *filp)
135 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
136 filp->f_path.dentry->d_parent->d_name.name,
137 filp->f_path.dentry->d_name.name);
139 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
141 /* Call generic open code in order to cache credentials */
142 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;
158 unsigned long gencount;
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, gencount;
183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __func__, (long long)desc->entry->cookie,
189 gencount = nfs_inc_attr_generation_counter();
190 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
191 NFS_SERVER(inode)->dtsize, desc->plus);
193 /* We requested READDIRPLUS, but the server doesn't grok it */
194 if (error == -ENOTSUPP && desc->plus) {
195 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
196 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
202 desc->timestamp = timestamp;
203 desc->gencount = gencount;
204 desc->timestamp_valid = 1;
205 SetPageUptodate(page);
206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
208 * through inode->i_mutex or some other mechanism.
210 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
222 int dir_decode(nfs_readdir_descriptor_t *desc)
224 __be32 *p = desc->ptr;
225 p = desc->decode(p, desc->entry, desc->plus);
229 if (desc->timestamp_valid) {
230 desc->entry->fattr->time_start = desc->timestamp;
231 desc->entry->fattr->gencount = desc->gencount;
233 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
238 void dir_page_release(nfs_readdir_descriptor_t *desc)
241 page_cache_release(desc->page);
247 * Given a pointer to a buffer that has already been filled by a call
248 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
250 * If the end of the buffer has been reached, return -EAGAIN, if not,
251 * return the offset within the buffer of the next entry to be
255 int find_dirent(nfs_readdir_descriptor_t *desc)
257 struct nfs_entry *entry = desc->entry;
261 while((status = dir_decode(desc)) == 0) {
262 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
263 __func__, (unsigned long long)entry->cookie);
264 if (entry->prev_cookie == *desc->dir_cookie)
266 if (loop_count++ > 200) {
275 * Given a pointer to a buffer that has already been filled by a call
276 * to readdir, find the entry at offset 'desc->file->f_pos'.
278 * If the end of the buffer has been reached, return -EAGAIN, if not,
279 * return the offset within the buffer of the next entry to be
283 int find_dirent_index(nfs_readdir_descriptor_t *desc)
285 struct nfs_entry *entry = desc->entry;
290 status = dir_decode(desc);
294 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
295 (unsigned long long)entry->cookie, desc->current_index);
297 if (desc->file->f_pos == desc->current_index) {
298 *desc->dir_cookie = entry->cookie;
301 desc->current_index++;
302 if (loop_count++ > 200) {
311 * Find the given page, and call find_dirent() or find_dirent_index in
312 * order to try to return the next entry.
315 int find_dirent_page(nfs_readdir_descriptor_t *desc)
317 struct inode *inode = desc->file->f_path.dentry->d_inode;
321 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
322 __func__, desc->page_index,
323 (long long) *desc->dir_cookie);
325 /* If we find the page in the page_cache, we cannot be sure
326 * how fresh the data is, so we will ignore readdir_plus attributes.
328 desc->timestamp_valid = 0;
329 page = read_cache_page(inode->i_mapping, desc->page_index,
330 (filler_t *)nfs_readdir_filler, desc);
332 status = PTR_ERR(page);
336 /* NOTE: Someone else may have changed the READDIRPLUS flag */
338 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
339 if (*desc->dir_cookie != 0)
340 status = find_dirent(desc);
342 status = find_dirent_index(desc);
344 dir_page_release(desc);
346 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
351 * Recurse through the page cache pages, and return a
352 * filled nfs_entry structure of the next directory entry if possible.
354 * The target for the search is '*desc->dir_cookie' if non-0,
355 * 'desc->file->f_pos' otherwise
358 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
363 /* Always search-by-index from the beginning of the cache */
364 if (*desc->dir_cookie == 0) {
365 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
366 (long long)desc->file->f_pos);
367 desc->page_index = 0;
368 desc->entry->cookie = desc->entry->prev_cookie = 0;
369 desc->entry->eof = 0;
370 desc->current_index = 0;
372 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
373 (unsigned long long)*desc->dir_cookie);
376 res = find_dirent_page(desc);
379 /* Align to beginning of next page */
381 if (loop_count++ > 200) {
387 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
391 static inline unsigned int dt_type(struct inode *inode)
393 return (inode->i_mode >> 12) & 15;
396 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
399 * Once we've found the start of the dirent within a page: fill 'er up...
402 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
405 struct file *file = desc->file;
406 struct nfs_entry *entry = desc->entry;
407 struct dentry *dentry = NULL;
412 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
413 (unsigned long long)entry->cookie);
416 unsigned d_type = DT_UNKNOWN;
417 /* Note: entry->prev_cookie contains the cookie for
418 * retrieving the current dirent on the server */
421 /* Get a dentry if we have one */
424 dentry = nfs_readdir_lookup(desc);
426 /* Use readdirplus info */
427 if (dentry != NULL && dentry->d_inode != NULL) {
428 d_type = dt_type(dentry->d_inode);
429 fileid = NFS_FILEID(dentry->d_inode);
432 res = filldir(dirent, entry->name, entry->len,
433 file->f_pos, nfs_compat_user_ino64(fileid),
438 *desc->dir_cookie = entry->cookie;
439 if (dir_decode(desc) != 0) {
443 if (loop_count++ > 200) {
448 dir_page_release(desc);
451 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
452 (unsigned long long)*desc->dir_cookie, res);
457 * If we cannot find a cookie in our cache, we suspect that this is
458 * because it points to a deleted file, so we ask the server to return
459 * whatever it thinks is the next entry. We then feed this to filldir.
460 * If all goes well, we should then be able to find our way round the
461 * cache on the next call to readdir_search_pagecache();
463 * NOTE: we cannot add the anonymous page to the pagecache because
464 * the data it contains might not be page aligned. Besides,
465 * we should already have a complete representation of the
466 * directory in the page cache by the time we get here.
469 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
472 struct file *file = desc->file;
473 struct inode *inode = file->f_path.dentry->d_inode;
474 struct rpc_cred *cred = nfs_file_cred(file);
475 struct page *page = NULL;
477 unsigned long timestamp, gencount;
479 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
480 (unsigned long long)*desc->dir_cookie);
482 page = alloc_page(GFP_HIGHUSER);
488 gencount = nfs_inc_attr_generation_counter();
489 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
490 *desc->dir_cookie, page,
491 NFS_SERVER(inode)->dtsize,
494 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
496 desc->timestamp = timestamp;
497 desc->gencount = gencount;
498 desc->timestamp_valid = 1;
499 if ((status = dir_decode(desc)) == 0)
500 desc->entry->prev_cookie = *desc->dir_cookie;
506 status = nfs_do_filldir(desc, dirent, filldir);
508 /* Reset read descriptor so it searches the page cache from
509 * the start upon the next call to readdir_search_pagecache() */
510 desc->page_index = 0;
511 desc->entry->cookie = desc->entry->prev_cookie = 0;
512 desc->entry->eof = 0;
514 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
518 dir_page_release(desc);
522 /* The file offset position represents the dirent entry number. A
523 last cookie cache takes care of the common case of reading the
526 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
528 struct dentry *dentry = filp->f_path.dentry;
529 struct inode *inode = dentry->d_inode;
530 nfs_readdir_descriptor_t my_desc,
532 struct nfs_entry my_entry;
534 struct nfs_fattr fattr;
537 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
538 dentry->d_parent->d_name.name, dentry->d_name.name,
539 (long long)filp->f_pos);
540 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
543 * filp->f_pos points to the dirent entry number.
544 * *desc->dir_cookie has the cookie for the next entry. We have
545 * to either find the entry with the appropriate number or
546 * revalidate the cookie.
548 memset(desc, 0, sizeof(*desc));
551 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
552 desc->decode = NFS_PROTO(inode)->decode_dirent;
553 desc->plus = NFS_USE_READDIRPLUS(inode);
555 my_entry.cookie = my_entry.prev_cookie = 0;
558 my_entry.fattr = &fattr;
559 nfs_fattr_init(&fattr);
560 desc->entry = &my_entry;
562 nfs_block_sillyrename(dentry);
563 res = nfs_revalidate_mapping(inode, filp->f_mapping);
567 while(!desc->entry->eof) {
568 res = readdir_search_pagecache(desc);
570 if (res == -EBADCOOKIE) {
571 /* This means either end of directory */
572 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
573 /* Or that the server has 'lost' a cookie */
574 res = uncached_readdir(desc, dirent, filldir);
581 if (res == -ETOOSMALL && desc->plus) {
582 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
583 nfs_zap_caches(inode);
585 desc->entry->eof = 0;
591 res = nfs_do_filldir(desc, dirent, filldir);
598 nfs_unblock_sillyrename(dentry);
601 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
602 dentry->d_parent->d_name.name, dentry->d_name.name,
607 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
609 struct dentry *dentry = filp->f_path.dentry;
610 struct inode *inode = dentry->d_inode;
612 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
613 dentry->d_parent->d_name.name,
617 mutex_lock(&inode->i_mutex);
620 offset += filp->f_pos;
628 if (offset != filp->f_pos) {
629 filp->f_pos = offset;
630 nfs_file_open_context(filp)->dir_cookie = 0;
633 mutex_unlock(&inode->i_mutex);
638 * All directory operations under NFS are synchronous, so fsync()
639 * is a dummy operation.
641 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
643 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
644 dentry->d_parent->d_name.name, dentry->d_name.name,
647 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
652 * nfs_force_lookup_revalidate - Mark the directory as having changed
653 * @dir - pointer to directory inode
655 * This forces the revalidation code in nfs_lookup_revalidate() to do a
656 * full lookup on all child dentries of 'dir' whenever a change occurs
657 * on the server that might have invalidated our dcache.
659 * The caller should be holding dir->i_lock
661 void nfs_force_lookup_revalidate(struct inode *dir)
663 NFS_I(dir)->cache_change_attribute++;
667 * A check for whether or not the parent directory has changed.
668 * In the case it has, we assume that the dentries are untrustworthy
669 * and may need to be looked up again.
671 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
675 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
677 if (!nfs_verify_change_attribute(dir, dentry->d_time))
679 /* Revalidate nfsi->cache_change_attribute before we declare a match */
680 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
682 if (!nfs_verify_change_attribute(dir, dentry->d_time))
688 * Return the intent data that applies to this particular path component
690 * Note that the current set of intents only apply to the very last
691 * component of the path.
692 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
694 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
696 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
698 return nd->flags & mask;
702 * Use intent information to check whether or not we're going to do
703 * an O_EXCL create using this path component.
705 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
707 if (NFS_PROTO(dir)->version == 2)
709 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
713 * Inode and filehandle revalidation for lookups.
715 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
716 * or if the intent information indicates that we're about to open this
717 * particular file and the "nocto" mount flag is not set.
721 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
723 struct nfs_server *server = NFS_SERVER(inode);
725 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
728 /* VFS wants an on-the-wire revalidation */
729 if (nd->flags & LOOKUP_REVAL)
731 /* This is an open(2) */
732 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
733 !(server->flags & NFS_MOUNT_NOCTO) &&
734 (S_ISREG(inode->i_mode) ||
735 S_ISDIR(inode->i_mode)))
739 return nfs_revalidate_inode(server, inode);
741 return __nfs_revalidate_inode(server, inode);
745 * We judge how long we want to trust negative
746 * dentries by looking at the parent inode mtime.
748 * If parent mtime has changed, we revalidate, else we wait for a
749 * period corresponding to the parent's attribute cache timeout value.
752 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
753 struct nameidata *nd)
755 /* Don't revalidate a negative dentry if we're creating a new file */
756 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
758 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
760 return !nfs_check_verifier(dir, dentry);
764 * This is called every time the dcache has a lookup hit,
765 * and we should check whether we can really trust that
768 * NOTE! The hit can be a negative hit too, don't assume
771 * If the parent directory is seen to have changed, we throw out the
772 * cached dentry and do a new lookup.
774 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
778 struct dentry *parent;
780 struct nfs_fh fhandle;
781 struct nfs_fattr fattr;
783 parent = dget_parent(dentry);
784 dir = parent->d_inode;
785 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
786 inode = dentry->d_inode;
789 if (nfs_neg_need_reval(dir, dentry, nd))
794 if (is_bad_inode(inode)) {
795 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
796 __func__, dentry->d_parent->d_name.name,
797 dentry->d_name.name);
801 if (nfs_have_delegation(inode, FMODE_READ))
802 goto out_set_verifier;
804 /* Force a full look up iff the parent directory has changed */
805 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
806 if (nfs_lookup_verify_inode(inode, nd))
811 if (NFS_STALE(inode))
814 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
817 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
819 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
823 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
826 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
827 __func__, dentry->d_parent->d_name.name,
828 dentry->d_name.name);
833 nfs_mark_for_revalidate(dir);
834 if (inode && S_ISDIR(inode->i_mode)) {
835 /* Purge readdir caches. */
836 nfs_zap_caches(inode);
837 /* If we have submounts, don't unhash ! */
838 if (have_submounts(dentry))
840 if (dentry->d_flags & DCACHE_DISCONNECTED)
842 shrink_dcache_parent(dentry);
846 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
847 __func__, dentry->d_parent->d_name.name,
848 dentry->d_name.name);
853 * This is called from dput() when d_count is going to 0.
855 static int nfs_dentry_delete(struct dentry *dentry)
857 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
858 dentry->d_parent->d_name.name, dentry->d_name.name,
861 /* Unhash any dentry with a stale inode */
862 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
865 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
866 /* Unhash it, so that ->d_iput() would be called */
869 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
870 /* Unhash it, so that ancestors of killed async unlink
871 * files will be cleaned up during umount */
878 static void nfs_drop_nlink(struct inode *inode)
880 spin_lock(&inode->i_lock);
881 if (inode->i_nlink > 0)
883 spin_unlock(&inode->i_lock);
887 * Called when the dentry loses inode.
888 * We use it to clean up silly-renamed files.
890 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
892 if (S_ISDIR(inode->i_mode))
893 /* drop any readdir cache as it could easily be old */
894 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
896 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
898 nfs_complete_unlink(dentry, inode);
903 const struct dentry_operations nfs_dentry_operations = {
904 .d_revalidate = nfs_lookup_revalidate,
905 .d_delete = nfs_dentry_delete,
906 .d_iput = nfs_dentry_iput,
909 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
912 struct dentry *parent;
913 struct inode *inode = NULL;
915 struct nfs_fh fhandle;
916 struct nfs_fattr fattr;
918 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
919 dentry->d_parent->d_name.name, dentry->d_name.name);
920 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
922 res = ERR_PTR(-ENAMETOOLONG);
923 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
926 res = ERR_PTR(-ENOMEM);
927 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
930 * If we're doing an exclusive create, optimize away the lookup
931 * but don't hash the dentry.
933 if (nfs_is_exclusive_create(dir, nd)) {
934 d_instantiate(dentry, NULL);
939 parent = dentry->d_parent;
940 /* Protect against concurrent sillydeletes */
941 nfs_block_sillyrename(parent);
942 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
943 if (error == -ENOENT)
946 res = ERR_PTR(error);
947 goto out_unblock_sillyrename;
949 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
950 res = (struct dentry *)inode;
952 goto out_unblock_sillyrename;
955 res = d_materialise_unique(dentry, inode);
958 goto out_unblock_sillyrename;
961 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
962 out_unblock_sillyrename:
963 nfs_unblock_sillyrename(parent);
969 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
971 const struct dentry_operations nfs4_dentry_operations = {
972 .d_revalidate = nfs_open_revalidate,
973 .d_delete = nfs_dentry_delete,
974 .d_iput = nfs_dentry_iput,
978 * Use intent information to determine whether we need to substitute
979 * the NFSv4-style stateful OPEN for the LOOKUP call
981 static int is_atomic_open(struct nameidata *nd)
983 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
985 /* NFS does not (yet) have a stateful open for directories */
986 if (nd->flags & LOOKUP_DIRECTORY)
988 /* Are we trying to write to a read only partition? */
989 if (__mnt_is_readonly(nd->path.mnt) &&
990 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
995 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
997 struct dentry *res = NULL;
1000 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1001 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1003 /* Check that we are indeed trying to open this file */
1004 if (!is_atomic_open(nd))
1007 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1008 res = ERR_PTR(-ENAMETOOLONG);
1011 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1013 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1015 if (nd->flags & LOOKUP_EXCL) {
1016 d_instantiate(dentry, NULL);
1020 /* Open the file on the server */
1021 res = nfs4_atomic_open(dir, dentry, nd);
1023 error = PTR_ERR(res);
1025 /* Make a negative dentry */
1029 /* This turned out not to be a regular file */
1034 if (!(nd->intent.open.flags & O_NOFOLLOW))
1040 } else if (res != NULL)
1045 return nfs_lookup(dir, dentry, nd);
1048 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1050 struct dentry *parent = NULL;
1051 struct inode *inode = dentry->d_inode;
1053 int openflags, ret = 0;
1055 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1057 parent = dget_parent(dentry);
1058 dir = parent->d_inode;
1059 /* We can't create new files in nfs_open_revalidate(), so we
1060 * optimize away revalidation of negative dentries.
1062 if (inode == NULL) {
1063 if (!nfs_neg_need_reval(dir, dentry, nd))
1068 /* NFS only supports OPEN on regular files */
1069 if (!S_ISREG(inode->i_mode))
1071 openflags = nd->intent.open.flags;
1072 /* We cannot do exclusive creation on a positive dentry */
1073 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1075 /* We can't create new files, or truncate existing ones here */
1076 openflags &= ~(O_CREAT|O_TRUNC);
1079 * Note: we're not holding inode->i_mutex and so may be racing with
1080 * operations that change the directory. We therefore save the
1081 * change attribute *before* we do the RPC call.
1083 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1092 return nfs_lookup_revalidate(dentry, nd);
1094 #endif /* CONFIG_NFSV4 */
1096 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1098 struct dentry *parent = desc->file->f_path.dentry;
1099 struct inode *dir = parent->d_inode;
1100 struct nfs_entry *entry = desc->entry;
1101 struct dentry *dentry, *alias;
1102 struct qstr name = {
1103 .name = entry->name,
1106 struct inode *inode;
1107 unsigned long verf = nfs_save_change_attribute(dir);
1111 if (name.name[0] == '.' && name.name[1] == '.')
1112 return dget_parent(parent);
1115 if (name.name[0] == '.')
1116 return dget(parent);
1119 spin_lock(&dir->i_lock);
1120 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1121 spin_unlock(&dir->i_lock);
1124 spin_unlock(&dir->i_lock);
1126 name.hash = full_name_hash(name.name, name.len);
1127 dentry = d_lookup(parent, &name);
1128 if (dentry != NULL) {
1129 /* Is this a positive dentry that matches the readdir info? */
1130 if (dentry->d_inode != NULL &&
1131 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1132 d_mountpoint(dentry))) {
1133 if (!desc->plus || entry->fh->size == 0)
1135 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1139 /* No, so d_drop to allow one to be created */
1143 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1145 if (name.len > NFS_SERVER(dir)->namelen)
1147 /* Note: caller is already holding the dir->i_mutex! */
1148 dentry = d_alloc(parent, &name);
1151 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1152 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1153 if (IS_ERR(inode)) {
1158 alias = d_materialise_unique(dentry, inode);
1159 if (alias != NULL) {
1167 nfs_set_verifier(dentry, verf);
1172 * Code common to create, mkdir, and mknod.
1174 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1175 struct nfs_fattr *fattr)
1177 struct dentry *parent = dget_parent(dentry);
1178 struct inode *dir = parent->d_inode;
1179 struct inode *inode;
1180 int error = -EACCES;
1184 /* We may have been initialized further down */
1185 if (dentry->d_inode)
1187 if (fhandle->size == 0) {
1188 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1192 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1193 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1194 struct nfs_server *server = NFS_SB(dentry->d_sb);
1195 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1199 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1200 error = PTR_ERR(inode);
1203 d_add(dentry, inode);
1208 nfs_mark_for_revalidate(dir);
1214 * Following a failed create operation, we drop the dentry rather
1215 * than retain a negative dentry. This avoids a problem in the event
1216 * that the operation succeeded on the server, but an error in the
1217 * reply path made it appear to have failed.
1219 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1220 struct nameidata *nd)
1226 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1227 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1229 attr.ia_mode = mode;
1230 attr.ia_valid = ATTR_MODE;
1232 if ((nd->flags & LOOKUP_CREATE) != 0)
1233 open_flags = nd->intent.open.flags;
1235 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1245 * See comments for nfs_proc_create regarding failed operations.
1248 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1253 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1254 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1256 if (!new_valid_dev(rdev))
1259 attr.ia_mode = mode;
1260 attr.ia_valid = ATTR_MODE;
1262 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1272 * See comments for nfs_proc_create regarding failed operations.
1274 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1279 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1280 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1282 attr.ia_valid = ATTR_MODE;
1283 attr.ia_mode = mode | S_IFDIR;
1285 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1294 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1296 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1300 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1304 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1305 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1307 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1308 /* Ensure the VFS deletes this inode */
1309 if (error == 0 && dentry->d_inode != NULL)
1310 clear_nlink(dentry->d_inode);
1311 else if (error == -ENOENT)
1312 nfs_dentry_handle_enoent(dentry);
1317 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1319 static unsigned int sillycounter;
1320 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1321 const int countersize = sizeof(sillycounter)*2;
1322 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1325 struct dentry *sdentry;
1328 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1329 dentry->d_parent->d_name.name, dentry->d_name.name,
1330 atomic_read(&dentry->d_count));
1331 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1334 * We don't allow a dentry to be silly-renamed twice.
1337 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1340 sprintf(silly, ".nfs%*.*Lx",
1341 fileidsize, fileidsize,
1342 (unsigned long long)NFS_FILEID(dentry->d_inode));
1344 /* Return delegation in anticipation of the rename */
1345 nfs_inode_return_delegation(dentry->d_inode);
1349 char *suffix = silly + slen - countersize;
1353 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1355 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1356 dentry->d_name.name, silly);
1358 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1360 * N.B. Better to return EBUSY here ... it could be
1361 * dangerous to delete the file while it's in use.
1363 if (IS_ERR(sdentry))
1365 } while(sdentry->d_inode != NULL); /* need negative lookup */
1367 qsilly.name = silly;
1368 qsilly.len = strlen(silly);
1369 if (dentry->d_inode) {
1370 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1372 nfs_mark_for_revalidate(dentry->d_inode);
1374 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1377 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1378 d_move(dentry, sdentry);
1379 error = nfs_async_unlink(dir, dentry);
1380 /* If we return 0 we don't unlink */
1388 * Remove a file after making sure there are no pending writes,
1389 * and after checking that the file has only one user.
1391 * We invalidate the attribute cache and free the inode prior to the operation
1392 * to avoid possible races if the server reuses the inode.
1394 static int nfs_safe_remove(struct dentry *dentry)
1396 struct inode *dir = dentry->d_parent->d_inode;
1397 struct inode *inode = dentry->d_inode;
1400 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1401 dentry->d_parent->d_name.name, dentry->d_name.name);
1403 /* If the dentry was sillyrenamed, we simply call d_delete() */
1404 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1409 if (inode != NULL) {
1410 nfs_inode_return_delegation(inode);
1411 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1412 /* The VFS may want to delete this inode */
1414 nfs_drop_nlink(inode);
1415 nfs_mark_for_revalidate(inode);
1417 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1418 if (error == -ENOENT)
1419 nfs_dentry_handle_enoent(dentry);
1424 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1425 * belongs to an active ".nfs..." file and we return -EBUSY.
1427 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1429 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1432 int need_rehash = 0;
1434 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1435 dir->i_ino, dentry->d_name.name);
1437 spin_lock(&dcache_lock);
1438 spin_lock(&dentry->d_lock);
1439 if (atomic_read(&dentry->d_count) > 1) {
1440 spin_unlock(&dentry->d_lock);
1441 spin_unlock(&dcache_lock);
1442 /* Start asynchronous writeout of the inode */
1443 write_inode_now(dentry->d_inode, 0);
1444 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)
1462 * To create a symbolic link, most file systems instantiate a new inode,
1463 * add a page to it containing the path, then write it out to the disk
1464 * using prepare_write/commit_write.
1466 * Unfortunately the NFS client can't create the in-core inode first
1467 * because it needs a file handle to create an in-core inode (see
1468 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1469 * symlink request has completed on the server.
1471 * So instead we allocate a raw page, copy the symname into it, then do
1472 * the SYMLINK request with the page as the buffer. If it succeeds, we
1473 * now have a new file handle and can instantiate an in-core NFS inode
1474 * and move the raw page into its mapping.
1476 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1478 struct pagevec lru_pvec;
1482 unsigned int pathlen = strlen(symname);
1485 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1486 dir->i_ino, dentry->d_name.name, symname);
1488 if (pathlen > PAGE_SIZE)
1489 return -ENAMETOOLONG;
1491 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1492 attr.ia_valid = ATTR_MODE;
1494 page = alloc_page(GFP_HIGHUSER);
1498 kaddr = kmap_atomic(page, KM_USER0);
1499 memcpy(kaddr, symname, pathlen);
1500 if (pathlen < PAGE_SIZE)
1501 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1502 kunmap_atomic(kaddr, KM_USER0);
1504 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1506 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1507 dir->i_sb->s_id, dir->i_ino,
1508 dentry->d_name.name, symname, error);
1515 * No big deal if we can't add this page to the page cache here.
1516 * READLINK will get the missing page from the server if needed.
1518 pagevec_init(&lru_pvec, 0);
1519 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1521 pagevec_add(&lru_pvec, page);
1522 pagevec_lru_add_file(&lru_pvec);
1523 SetPageUptodate(page);
1532 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1534 struct inode *inode = old_dentry->d_inode;
1537 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1538 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1539 dentry->d_parent->d_name.name, dentry->d_name.name);
1541 nfs_inode_return_delegation(inode);
1544 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1546 atomic_inc(&inode->i_count);
1547 d_add(dentry, inode);
1554 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1555 * different file handle for the same inode after a rename (e.g. when
1556 * moving to a different directory). A fail-safe method to do so would
1557 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1558 * rename the old file using the sillyrename stuff. This way, the original
1559 * file in old_dir will go away when the last process iput()s the inode.
1563 * It actually works quite well. One needs to have the possibility for
1564 * at least one ".nfs..." file in each directory the file ever gets
1565 * moved or linked to which happens automagically with the new
1566 * implementation that only depends on the dcache stuff instead of
1567 * using the inode layer
1569 * Unfortunately, things are a little more complicated than indicated
1570 * above. For a cross-directory move, we want to make sure we can get
1571 * rid of the old inode after the operation. This means there must be
1572 * no pending writes (if it's a file), and the use count must be 1.
1573 * If these conditions are met, we can drop the dentries before doing
1576 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1577 struct inode *new_dir, struct dentry *new_dentry)
1579 struct inode *old_inode = old_dentry->d_inode;
1580 struct inode *new_inode = new_dentry->d_inode;
1581 struct dentry *dentry = NULL, *rehash = NULL;
1584 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1585 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1586 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1587 atomic_read(&new_dentry->d_count));
1590 * For non-directories, check whether the target is busy and if so,
1591 * make a copy of the dentry and then do a silly-rename. If the
1592 * silly-rename succeeds, the copied dentry is hashed and becomes
1595 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1597 * To prevent any new references to the target during the
1598 * rename, we unhash the dentry in advance.
1600 if (!d_unhashed(new_dentry)) {
1602 rehash = new_dentry;
1605 if (atomic_read(&new_dentry->d_count) > 2) {
1608 /* copy the target dentry's name */
1609 dentry = d_alloc(new_dentry->d_parent,
1610 &new_dentry->d_name);
1614 /* silly-rename the existing target ... */
1615 err = nfs_sillyrename(new_dir, new_dentry);
1619 new_dentry = dentry;
1626 * ... prune child dentries and writebacks if needed.
1628 if (atomic_read(&old_dentry->d_count) > 1) {
1629 if (S_ISREG(old_inode->i_mode))
1630 nfs_wb_all(old_inode);
1631 shrink_dcache_parent(old_dentry);
1633 nfs_inode_return_delegation(old_inode);
1635 if (new_inode != NULL)
1636 nfs_inode_return_delegation(new_inode);
1638 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1639 new_dir, &new_dentry->d_name);
1640 nfs_mark_for_revalidate(old_inode);
1645 if (new_inode != NULL)
1646 nfs_drop_nlink(new_inode);
1647 d_move(old_dentry, new_dentry);
1648 nfs_set_verifier(new_dentry,
1649 nfs_save_change_attribute(new_dir));
1650 } else if (error == -ENOENT)
1651 nfs_dentry_handle_enoent(old_dentry);
1653 /* new dentry created? */
1659 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1660 static LIST_HEAD(nfs_access_lru_list);
1661 static atomic_long_t nfs_access_nr_entries;
1663 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1665 put_rpccred(entry->cred);
1667 smp_mb__before_atomic_dec();
1668 atomic_long_dec(&nfs_access_nr_entries);
1669 smp_mb__after_atomic_dec();
1672 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1675 struct nfs_inode *nfsi;
1676 struct nfs_access_entry *cache;
1679 spin_lock(&nfs_access_lru_lock);
1680 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1681 struct rw_semaphore *s_umount;
1682 struct inode *inode;
1684 if (nr_to_scan-- == 0)
1686 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1687 if (!down_read_trylock(s_umount))
1689 inode = igrab(&nfsi->vfs_inode);
1690 if (inode == NULL) {
1694 spin_lock(&inode->i_lock);
1695 if (list_empty(&nfsi->access_cache_entry_lru))
1696 goto remove_lru_entry;
1697 cache = list_entry(nfsi->access_cache_entry_lru.next,
1698 struct nfs_access_entry, lru);
1699 list_move(&cache->lru, &head);
1700 rb_erase(&cache->rb_node, &nfsi->access_cache);
1701 if (!list_empty(&nfsi->access_cache_entry_lru))
1702 list_move_tail(&nfsi->access_cache_inode_lru,
1703 &nfs_access_lru_list);
1706 list_del_init(&nfsi->access_cache_inode_lru);
1707 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1709 spin_unlock(&inode->i_lock);
1710 spin_unlock(&nfs_access_lru_lock);
1715 spin_unlock(&nfs_access_lru_lock);
1716 while (!list_empty(&head)) {
1717 cache = list_entry(head.next, struct nfs_access_entry, lru);
1718 list_del(&cache->lru);
1719 nfs_access_free_entry(cache);
1721 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1724 static void __nfs_access_zap_cache(struct inode *inode)
1726 struct nfs_inode *nfsi = NFS_I(inode);
1727 struct rb_root *root_node = &nfsi->access_cache;
1728 struct rb_node *n, *dispose = NULL;
1729 struct nfs_access_entry *entry;
1731 /* Unhook entries from the cache */
1732 while ((n = rb_first(root_node)) != NULL) {
1733 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1734 rb_erase(n, root_node);
1735 list_del(&entry->lru);
1736 n->rb_left = dispose;
1739 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1740 spin_unlock(&inode->i_lock);
1742 /* Now kill them all! */
1743 while (dispose != NULL) {
1745 dispose = n->rb_left;
1746 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1750 void nfs_access_zap_cache(struct inode *inode)
1752 /* Remove from global LRU init */
1753 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1754 spin_lock(&nfs_access_lru_lock);
1755 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1756 spin_unlock(&nfs_access_lru_lock);
1759 spin_lock(&inode->i_lock);
1760 /* This will release the spinlock */
1761 __nfs_access_zap_cache(inode);
1764 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1766 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1767 struct nfs_access_entry *entry;
1770 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1772 if (cred < entry->cred)
1774 else if (cred > entry->cred)
1782 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1784 struct nfs_inode *nfsi = NFS_I(inode);
1785 struct nfs_access_entry *cache;
1788 spin_lock(&inode->i_lock);
1789 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1791 cache = nfs_access_search_rbtree(inode, cred);
1794 if (!nfs_have_delegated_attributes(inode) &&
1795 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1797 res->jiffies = cache->jiffies;
1798 res->cred = cache->cred;
1799 res->mask = cache->mask;
1800 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1803 spin_unlock(&inode->i_lock);
1806 rb_erase(&cache->rb_node, &nfsi->access_cache);
1807 list_del(&cache->lru);
1808 spin_unlock(&inode->i_lock);
1809 nfs_access_free_entry(cache);
1812 /* This will release the spinlock */
1813 __nfs_access_zap_cache(inode);
1817 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1819 struct nfs_inode *nfsi = NFS_I(inode);
1820 struct rb_root *root_node = &nfsi->access_cache;
1821 struct rb_node **p = &root_node->rb_node;
1822 struct rb_node *parent = NULL;
1823 struct nfs_access_entry *entry;
1825 spin_lock(&inode->i_lock);
1826 while (*p != NULL) {
1828 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1830 if (set->cred < entry->cred)
1831 p = &parent->rb_left;
1832 else if (set->cred > entry->cred)
1833 p = &parent->rb_right;
1837 rb_link_node(&set->rb_node, parent, p);
1838 rb_insert_color(&set->rb_node, root_node);
1839 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1840 spin_unlock(&inode->i_lock);
1843 rb_replace_node(parent, &set->rb_node, root_node);
1844 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1845 list_del(&entry->lru);
1846 spin_unlock(&inode->i_lock);
1847 nfs_access_free_entry(entry);
1850 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1852 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1855 RB_CLEAR_NODE(&cache->rb_node);
1856 cache->jiffies = set->jiffies;
1857 cache->cred = get_rpccred(set->cred);
1858 cache->mask = set->mask;
1860 nfs_access_add_rbtree(inode, cache);
1862 /* Update accounting */
1863 smp_mb__before_atomic_inc();
1864 atomic_long_inc(&nfs_access_nr_entries);
1865 smp_mb__after_atomic_inc();
1867 /* Add inode to global LRU list */
1868 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1869 spin_lock(&nfs_access_lru_lock);
1870 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1871 spin_unlock(&nfs_access_lru_lock);
1875 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1877 struct nfs_access_entry cache;
1880 status = nfs_access_get_cached(inode, cred, &cache);
1884 /* Be clever: ask server to check for all possible rights */
1885 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1887 cache.jiffies = jiffies;
1888 status = NFS_PROTO(inode)->access(inode, &cache);
1890 if (status == -ESTALE) {
1891 nfs_zap_caches(inode);
1892 if (!S_ISDIR(inode->i_mode))
1893 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1897 nfs_access_add_cache(inode, &cache);
1899 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1904 static int nfs_open_permission_mask(int openflags)
1908 if (openflags & FMODE_READ)
1910 if (openflags & FMODE_WRITE)
1912 if (openflags & FMODE_EXEC)
1917 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1919 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1922 int nfs_permission(struct inode *inode, int mask)
1924 struct rpc_cred *cred;
1927 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1929 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1931 /* Is this sys_access() ? */
1932 if (mask & MAY_ACCESS)
1935 switch (inode->i_mode & S_IFMT) {
1939 /* NFSv4 has atomic_open... */
1940 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1941 && (mask & MAY_OPEN)
1942 && !(mask & MAY_EXEC))
1947 * Optimize away all write operations, since the server
1948 * will check permissions when we perform the op.
1950 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1955 if (!NFS_PROTO(inode)->access)
1958 cred = rpc_lookup_cred();
1959 if (!IS_ERR(cred)) {
1960 res = nfs_do_access(inode, cred, mask);
1963 res = PTR_ERR(cred);
1965 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
1968 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1969 inode->i_sb->s_id, inode->i_ino, mask, res);
1972 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1974 res = generic_permission(inode, mask, NULL);
1980 * version-control: t
1981 * kept-new-versions: 5