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"
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(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
139 /* Call generic open code in order to cache credentials */
140 res = nfs_open(inode, filp);
145 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
149 unsigned long page_index;
152 loff_t current_index;
153 struct nfs_entry *entry;
154 decode_dirent_t decode;
157 unsigned long timestamp;
159 } nfs_readdir_descriptor_t;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176 struct file *file = desc->file;
177 struct inode *inode = file->f_path.dentry->d_inode;
178 struct rpc_cred *cred = nfs_file_cred(file);
179 unsigned long timestamp;
182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__, (long long)desc->entry->cookie,
188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
189 NFS_SERVER(inode)->dtsize, desc->plus);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error == -ENOTSUPP && desc->plus) {
193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
200 desc->timestamp = timestamp;
201 desc->timestamp_valid = 1;
202 SetPageUptodate(page);
203 spin_lock(&inode->i_lock);
204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
205 spin_unlock(&inode->i_lock);
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 (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
219 nfs_zap_caches(inode);
225 int dir_decode(nfs_readdir_descriptor_t *desc)
227 __be32 *p = desc->ptr;
228 p = desc->decode(p, desc->entry, desc->plus);
232 if (desc->timestamp_valid)
233 desc->entry->fattr->time_start = desc->timestamp;
235 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
240 void dir_page_release(nfs_readdir_descriptor_t *desc)
243 page_cache_release(desc->page);
249 * Given a pointer to a buffer that has already been filled by a call
250 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
252 * If the end of the buffer has been reached, return -EAGAIN, if not,
253 * return the offset within the buffer of the next entry to be
257 int find_dirent(nfs_readdir_descriptor_t *desc)
259 struct nfs_entry *entry = desc->entry;
263 while((status = dir_decode(desc)) == 0) {
264 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
265 __FUNCTION__, (unsigned long long)entry->cookie);
266 if (entry->prev_cookie == *desc->dir_cookie)
268 if (loop_count++ > 200) {
277 * Given a pointer to a buffer that has already been filled by a call
278 * to readdir, find the entry at offset 'desc->file->f_pos'.
280 * If the end of the buffer has been reached, return -EAGAIN, if not,
281 * return the offset within the buffer of the next entry to be
285 int find_dirent_index(nfs_readdir_descriptor_t *desc)
287 struct nfs_entry *entry = desc->entry;
292 status = dir_decode(desc);
296 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
297 (unsigned long long)entry->cookie, desc->current_index);
299 if (desc->file->f_pos == desc->current_index) {
300 *desc->dir_cookie = entry->cookie;
303 desc->current_index++;
304 if (loop_count++ > 200) {
313 * Find the given page, and call find_dirent() or find_dirent_index in
314 * order to try to return the next entry.
317 int find_dirent_page(nfs_readdir_descriptor_t *desc)
319 struct inode *inode = desc->file->f_path.dentry->d_inode;
323 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
324 __FUNCTION__, desc->page_index,
325 (long long) *desc->dir_cookie);
327 /* If we find the page in the page_cache, we cannot be sure
328 * how fresh the data is, so we will ignore readdir_plus attributes.
330 desc->timestamp_valid = 0;
331 page = read_cache_page(inode->i_mapping, desc->page_index,
332 (filler_t *)nfs_readdir_filler, desc);
334 status = PTR_ERR(page);
338 /* NOTE: Someone else may have changed the READDIRPLUS flag */
340 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
341 if (*desc->dir_cookie != 0)
342 status = find_dirent(desc);
344 status = find_dirent_index(desc);
346 dir_page_release(desc);
348 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
353 * Recurse through the page cache pages, and return a
354 * filled nfs_entry structure of the next directory entry if possible.
356 * The target for the search is '*desc->dir_cookie' if non-0,
357 * 'desc->file->f_pos' otherwise
360 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
365 /* Always search-by-index from the beginning of the cache */
366 if (*desc->dir_cookie == 0) {
367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
368 (long long)desc->file->f_pos);
369 desc->page_index = 0;
370 desc->entry->cookie = desc->entry->prev_cookie = 0;
371 desc->entry->eof = 0;
372 desc->current_index = 0;
374 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
375 (unsigned long long)*desc->dir_cookie);
378 res = find_dirent_page(desc);
381 /* Align to beginning of next page */
383 if (loop_count++ > 200) {
389 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
393 static inline unsigned int dt_type(struct inode *inode)
395 return (inode->i_mode >> 12) & 15;
398 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
401 * Once we've found the start of the dirent within a page: fill 'er up...
404 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
407 struct file *file = desc->file;
408 struct nfs_entry *entry = desc->entry;
409 struct dentry *dentry = NULL;
414 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
415 (unsigned long long)entry->cookie);
418 unsigned d_type = DT_UNKNOWN;
419 /* Note: entry->prev_cookie contains the cookie for
420 * retrieving the current dirent on the server */
423 /* Get a dentry if we have one */
426 dentry = nfs_readdir_lookup(desc);
428 /* Use readdirplus info */
429 if (dentry != NULL && dentry->d_inode != NULL) {
430 d_type = dt_type(dentry->d_inode);
431 fileid = NFS_FILEID(dentry->d_inode);
434 res = filldir(dirent, entry->name, entry->len,
435 file->f_pos, fileid, d_type);
439 *desc->dir_cookie = entry->cookie;
440 if (dir_decode(desc) != 0) {
444 if (loop_count++ > 200) {
449 dir_page_release(desc);
452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
453 (unsigned long long)*desc->dir_cookie, res);
458 * If we cannot find a cookie in our cache, we suspect that this is
459 * because it points to a deleted file, so we ask the server to return
460 * whatever it thinks is the next entry. We then feed this to filldir.
461 * If all goes well, we should then be able to find our way round the
462 * cache on the next call to readdir_search_pagecache();
464 * NOTE: we cannot add the anonymous page to the pagecache because
465 * the data it contains might not be page aligned. Besides,
466 * we should already have a complete representation of the
467 * directory in the page cache by the time we get here.
470 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
473 struct file *file = desc->file;
474 struct inode *inode = file->f_path.dentry->d_inode;
475 struct rpc_cred *cred = nfs_file_cred(file);
476 struct page *page = NULL;
478 unsigned long timestamp;
480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
481 (unsigned long long)*desc->dir_cookie);
483 page = alloc_page(GFP_HIGHUSER);
489 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
491 NFS_SERVER(inode)->dtsize,
493 spin_lock(&inode->i_lock);
494 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
495 spin_unlock(&inode->i_lock);
497 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
498 if (desc->error >= 0) {
499 desc->timestamp = timestamp;
500 desc->timestamp_valid = 1;
501 if ((status = dir_decode(desc)) == 0)
502 desc->entry->prev_cookie = *desc->dir_cookie;
508 status = nfs_do_filldir(desc, dirent, filldir);
510 /* Reset read descriptor so it searches the page cache from
511 * the start upon the next call to readdir_search_pagecache() */
512 desc->page_index = 0;
513 desc->entry->cookie = desc->entry->prev_cookie = 0;
514 desc->entry->eof = 0;
516 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
517 __FUNCTION__, status);
520 dir_page_release(desc);
524 /* The file offset position represents the dirent entry number. A
525 last cookie cache takes care of the common case of reading the
528 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
530 struct dentry *dentry = filp->f_path.dentry;
531 struct inode *inode = dentry->d_inode;
532 nfs_readdir_descriptor_t my_desc,
534 struct nfs_entry my_entry;
536 struct nfs_fattr fattr;
539 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
540 dentry->d_parent->d_name.name, dentry->d_name.name,
541 (long long)filp->f_pos);
542 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
546 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
553 * filp->f_pos points to the dirent entry number.
554 * *desc->dir_cookie has the cookie for the next entry. We have
555 * to either find the entry with the appropriate number or
556 * revalidate the cookie.
558 memset(desc, 0, sizeof(*desc));
561 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
562 desc->decode = NFS_PROTO(inode)->decode_dirent;
563 desc->plus = NFS_USE_READDIRPLUS(inode);
565 my_entry.cookie = my_entry.prev_cookie = 0;
568 my_entry.fattr = &fattr;
569 nfs_fattr_init(&fattr);
570 desc->entry = &my_entry;
572 while(!desc->entry->eof) {
573 res = readdir_search_pagecache(desc);
575 if (res == -EBADCOOKIE) {
576 /* This means either end of directory */
577 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
578 /* Or that the server has 'lost' a cookie */
579 res = uncached_readdir(desc, dirent, filldir);
586 if (res == -ETOOSMALL && desc->plus) {
587 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
588 nfs_zap_caches(inode);
590 desc->entry->eof = 0;
596 res = nfs_do_filldir(desc, dirent, filldir);
605 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
606 dentry->d_parent->d_name.name, dentry->d_name.name,
611 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
613 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
616 offset += filp->f_pos;
624 if (offset != filp->f_pos) {
625 filp->f_pos = offset;
626 nfs_file_open_context(filp)->dir_cookie = 0;
629 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
634 * All directory operations under NFS are synchronous, so fsync()
635 * is a dummy operation.
637 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
639 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
640 dentry->d_parent->d_name.name, dentry->d_name.name,
647 * A check for whether or not the parent directory has changed.
648 * In the case it has, we assume that the dentries are untrustworthy
649 * and may need to be looked up again.
651 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
657 verf = dentry->d_time;
658 if (nfs_caches_unstable(dir)
659 || verf != NFS_I(dir)->cache_change_attribute)
664 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
666 dentry->d_time = verf;
670 * Return the intent data that applies to this particular path component
672 * Note that the current set of intents only apply to the very last
673 * component of the path.
674 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
676 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
678 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
680 return nd->flags & mask;
684 * Inode and filehandle revalidation for lookups.
686 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
687 * or if the intent information indicates that we're about to open this
688 * particular file and the "nocto" mount flag is not set.
692 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
694 struct nfs_server *server = NFS_SERVER(inode);
697 /* VFS wants an on-the-wire revalidation */
698 if (nd->flags & LOOKUP_REVAL)
700 /* This is an open(2) */
701 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
702 !(server->flags & NFS_MOUNT_NOCTO) &&
703 (S_ISREG(inode->i_mode) ||
704 S_ISDIR(inode->i_mode)))
707 return nfs_revalidate_inode(server, inode);
709 return __nfs_revalidate_inode(server, inode);
713 * We judge how long we want to trust negative
714 * dentries by looking at the parent inode mtime.
716 * If parent mtime has changed, we revalidate, else we wait for a
717 * period corresponding to the parent's attribute cache timeout value.
720 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
721 struct nameidata *nd)
723 /* Don't revalidate a negative dentry if we're creating a new file */
724 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
726 return !nfs_check_verifier(dir, dentry);
730 * This is called every time the dcache has a lookup hit,
731 * and we should check whether we can really trust that
734 * NOTE! The hit can be a negative hit too, don't assume
737 * If the parent directory is seen to have changed, we throw out the
738 * cached dentry and do a new lookup.
740 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
744 struct dentry *parent;
746 struct nfs_fh fhandle;
747 struct nfs_fattr fattr;
748 unsigned long verifier;
750 parent = dget_parent(dentry);
752 dir = parent->d_inode;
753 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
754 inode = dentry->d_inode;
756 /* Revalidate parent directory attribute cache */
757 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
761 if (nfs_neg_need_reval(dir, dentry, nd))
766 if (is_bad_inode(inode)) {
767 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
768 __FUNCTION__, dentry->d_parent->d_name.name,
769 dentry->d_name.name);
773 /* Force a full look up iff the parent directory has changed */
774 if (nfs_check_verifier(dir, dentry)) {
775 if (nfs_lookup_verify_inode(inode, nd))
780 if (NFS_STALE(inode))
783 verifier = nfs_save_change_attribute(dir);
784 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
787 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
789 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
792 nfs_set_verifier(dentry, verifier);
796 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
797 __FUNCTION__, dentry->d_parent->d_name.name,
798 dentry->d_name.name);
804 if (inode && S_ISDIR(inode->i_mode)) {
805 /* Purge readdir caches. */
806 nfs_zap_caches(inode);
807 /* If we have submounts, don't unhash ! */
808 if (have_submounts(dentry))
810 shrink_dcache_parent(dentry);
815 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
816 __FUNCTION__, dentry->d_parent->d_name.name,
817 dentry->d_name.name);
822 * This is called from dput() when d_count is going to 0.
824 static int nfs_dentry_delete(struct dentry *dentry)
826 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
827 dentry->d_parent->d_name.name, dentry->d_name.name,
830 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
831 /* Unhash it, so that ->d_iput() would be called */
834 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
835 /* Unhash it, so that ancestors of killed async unlink
836 * files will be cleaned up during umount */
844 * Called when the dentry loses inode.
845 * We use it to clean up silly-renamed files.
847 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
849 nfs_inode_return_delegation(inode);
850 if (S_ISDIR(inode->i_mode))
851 /* drop any readdir cache as it could easily be old */
852 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
854 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
857 nfs_complete_unlink(dentry, inode);
863 struct dentry_operations nfs_dentry_operations = {
864 .d_revalidate = nfs_lookup_revalidate,
865 .d_delete = nfs_dentry_delete,
866 .d_iput = nfs_dentry_iput,
870 * Use intent information to check whether or not we're going to do
871 * an O_EXCL create using this path component.
874 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
876 if (NFS_PROTO(dir)->version == 2)
878 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
880 return (nd->intent.open.flags & O_EXCL) != 0;
883 static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
885 struct nfs_server *server = NFS_SERVER(dir);
887 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
888 /* Revalidate fsid using the parent directory */
889 return __nfs_revalidate_inode(server, dir);
893 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
896 struct inode *inode = NULL;
898 struct nfs_fh fhandle;
899 struct nfs_fattr fattr;
901 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
902 dentry->d_parent->d_name.name, dentry->d_name.name);
903 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
905 res = ERR_PTR(-ENAMETOOLONG);
906 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
909 res = ERR_PTR(-ENOMEM);
910 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
915 * If we're doing an exclusive create, optimize away the lookup
916 * but don't hash the dentry.
918 if (nfs_is_exclusive_create(dir, nd)) {
919 d_instantiate(dentry, NULL);
924 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
925 if (error == -ENOENT)
928 res = ERR_PTR(error);
931 error = nfs_reval_fsid(dir, &fattr);
933 res = ERR_PTR(error);
936 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
937 res = (struct dentry *)inode;
942 res = d_materialise_unique(dentry, inode);
944 struct dentry *parent;
947 /* Was a directory renamed! */
948 parent = dget_parent(res);
949 if (!IS_ROOT(parent))
950 nfs_mark_for_revalidate(parent->d_inode);
954 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
962 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
964 struct dentry_operations nfs4_dentry_operations = {
965 .d_revalidate = nfs_open_revalidate,
966 .d_delete = nfs_dentry_delete,
967 .d_iput = nfs_dentry_iput,
971 * Use intent information to determine whether we need to substitute
972 * the NFSv4-style stateful OPEN for the LOOKUP call
974 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
976 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
978 /* NFS does not (yet) have a stateful open for directories */
979 if (nd->flags & LOOKUP_DIRECTORY)
981 /* Are we trying to write to a read only partition? */
982 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
987 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
989 struct dentry *res = NULL;
992 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
993 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
995 /* Check that we are indeed trying to open this file */
996 if (!is_atomic_open(dir, nd))
999 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1000 res = ERR_PTR(-ENAMETOOLONG);
1003 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1005 /* Let vfs_create() deal with O_EXCL */
1006 if (nd->intent.open.flags & O_EXCL) {
1007 d_add(dentry, NULL);
1011 /* Open the file on the server */
1013 /* Revalidate parent directory attribute cache */
1014 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1016 res = ERR_PTR(error);
1021 if (nd->intent.open.flags & O_CREAT) {
1022 nfs_begin_data_update(dir);
1023 res = nfs4_atomic_open(dir, dentry, nd);
1024 nfs_end_data_update(dir);
1026 res = nfs4_atomic_open(dir, dentry, nd);
1029 error = PTR_ERR(res);
1031 /* Make a negative dentry */
1035 /* This turned out not to be a regular file */
1040 if (!(nd->intent.open.flags & O_NOFOLLOW))
1046 } else if (res != NULL)
1048 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1052 return nfs_lookup(dir, dentry, nd);
1055 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1057 struct dentry *parent = NULL;
1058 struct inode *inode = dentry->d_inode;
1060 unsigned long verifier;
1061 int openflags, ret = 0;
1063 parent = dget_parent(dentry);
1064 dir = parent->d_inode;
1065 if (!is_atomic_open(dir, nd))
1067 /* We can't create new files in nfs_open_revalidate(), so we
1068 * optimize away revalidation of negative dentries.
1072 /* NFS only supports OPEN on regular files */
1073 if (!S_ISREG(inode->i_mode))
1075 openflags = nd->intent.open.flags;
1076 /* We cannot do exclusive creation on a positive dentry */
1077 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1079 /* We can't create new files, or truncate existing ones here */
1080 openflags &= ~(O_CREAT|O_TRUNC);
1083 * Note: we're not holding inode->i_mutex and so may be racing with
1084 * operations that change the directory. We therefore save the
1085 * change attribute *before* we do the RPC call.
1088 verifier = nfs_save_change_attribute(dir);
1089 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1091 nfs_set_verifier(dentry, verifier);
1100 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1102 return nfs_lookup_revalidate(dentry, nd);
1104 #endif /* CONFIG_NFSV4 */
1106 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1108 struct dentry *parent = desc->file->f_path.dentry;
1109 struct inode *dir = parent->d_inode;
1110 struct nfs_entry *entry = desc->entry;
1111 struct dentry *dentry, *alias;
1112 struct qstr name = {
1113 .name = entry->name,
1116 struct inode *inode;
1120 if (name.name[0] == '.' && name.name[1] == '.')
1121 return dget_parent(parent);
1124 if (name.name[0] == '.')
1125 return dget(parent);
1127 name.hash = full_name_hash(name.name, name.len);
1128 dentry = d_lookup(parent, &name);
1129 if (dentry != NULL) {
1130 /* Is this a positive dentry that matches the readdir info? */
1131 if (dentry->d_inode != NULL &&
1132 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1133 d_mountpoint(dentry))) {
1134 if (!desc->plus || entry->fh->size == 0)
1136 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1140 /* No, so d_drop to allow one to be created */
1144 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1146 if (name.len > NFS_SERVER(dir)->namelen)
1148 /* Note: caller is already holding the dir->i_mutex! */
1149 dentry = d_alloc(parent, &name);
1152 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1153 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1154 if (IS_ERR(inode)) {
1159 alias = d_materialise_unique(dentry, inode);
1160 if (alias != NULL) {
1168 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1173 * Code common to create, mkdir, and mknod.
1175 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1176 struct nfs_fattr *fattr)
1178 struct inode *inode;
1179 int error = -EACCES;
1181 /* We may have been initialized further down */
1182 if (dentry->d_inode)
1184 if (fhandle->size == 0) {
1185 struct inode *dir = dentry->d_parent->d_inode;
1186 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1190 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1191 struct nfs_server *server = NFS_SB(dentry->d_sb);
1192 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1196 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1197 error = PTR_ERR(inode);
1200 d_instantiate(dentry, inode);
1201 if (d_unhashed(dentry))
1207 * Following a failed create operation, we drop the dentry rather
1208 * than retain a negative dentry. This avoids a problem in the event
1209 * that the operation succeeded on the server, but an error in the
1210 * reply path made it appear to have failed.
1212 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1213 struct nameidata *nd)
1219 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1220 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1222 attr.ia_mode = mode;
1223 attr.ia_valid = ATTR_MODE;
1225 if ((nd->flags & LOOKUP_CREATE) != 0)
1226 open_flags = nd->intent.open.flags;
1229 nfs_begin_data_update(dir);
1230 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1231 nfs_end_data_update(dir);
1234 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1244 * See comments for nfs_proc_create regarding failed operations.
1247 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1252 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1253 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1255 if (!new_valid_dev(rdev))
1258 attr.ia_mode = mode;
1259 attr.ia_valid = ATTR_MODE;
1262 nfs_begin_data_update(dir);
1263 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1264 nfs_end_data_update(dir);
1267 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1277 * See comments for nfs_proc_create regarding failed operations.
1279 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1284 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1285 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1287 attr.ia_valid = ATTR_MODE;
1288 attr.ia_mode = mode | S_IFDIR;
1291 nfs_begin_data_update(dir);
1292 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1293 nfs_end_data_update(dir);
1296 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1305 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1309 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1310 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1313 nfs_begin_data_update(dir);
1314 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1315 /* Ensure the VFS deletes this inode */
1316 if (error == 0 && dentry->d_inode != NULL)
1317 clear_nlink(dentry->d_inode);
1318 nfs_end_data_update(dir);
1324 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1326 static unsigned int sillycounter;
1327 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1328 const int countersize = sizeof(sillycounter)*2;
1329 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1332 struct dentry *sdentry;
1335 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1336 dentry->d_parent->d_name.name, dentry->d_name.name,
1337 atomic_read(&dentry->d_count));
1338 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1341 * We don't allow a dentry to be silly-renamed twice.
1344 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1347 sprintf(silly, ".nfs%*.*Lx",
1348 fileidsize, fileidsize,
1349 (unsigned long long)NFS_FILEID(dentry->d_inode));
1351 /* Return delegation in anticipation of the rename */
1352 nfs_inode_return_delegation(dentry->d_inode);
1356 char *suffix = silly + slen - countersize;
1360 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1362 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1363 dentry->d_name.name, silly);
1365 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1367 * N.B. Better to return EBUSY here ... it could be
1368 * dangerous to delete the file while it's in use.
1370 if (IS_ERR(sdentry))
1372 } while(sdentry->d_inode != NULL); /* need negative lookup */
1374 qsilly.name = silly;
1375 qsilly.len = strlen(silly);
1376 nfs_begin_data_update(dir);
1377 if (dentry->d_inode) {
1378 nfs_begin_data_update(dentry->d_inode);
1379 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1381 nfs_mark_for_revalidate(dentry->d_inode);
1382 nfs_end_data_update(dentry->d_inode);
1384 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1386 nfs_end_data_update(dir);
1388 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1389 d_move(dentry, sdentry);
1390 error = nfs_async_unlink(dir, dentry);
1391 /* If we return 0 we don't unlink */
1399 * Remove a file after making sure there are no pending writes,
1400 * and after checking that the file has only one user.
1402 * We invalidate the attribute cache and free the inode prior to the operation
1403 * to avoid possible races if the server reuses the inode.
1405 static int nfs_safe_remove(struct dentry *dentry)
1407 struct inode *dir = dentry->d_parent->d_inode;
1408 struct inode *inode = dentry->d_inode;
1411 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1412 dentry->d_parent->d_name.name, dentry->d_name.name);
1414 /* If the dentry was sillyrenamed, we simply call d_delete() */
1415 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1420 nfs_begin_data_update(dir);
1421 if (inode != NULL) {
1422 nfs_inode_return_delegation(inode);
1423 nfs_begin_data_update(inode);
1424 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1425 /* The VFS may want to delete this inode */
1428 nfs_mark_for_revalidate(inode);
1429 nfs_end_data_update(inode);
1431 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1432 nfs_end_data_update(dir);
1437 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1438 * belongs to an active ".nfs..." file and we return -EBUSY.
1440 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1442 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1445 int need_rehash = 0;
1447 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1448 dir->i_ino, dentry->d_name.name);
1451 spin_lock(&dcache_lock);
1452 spin_lock(&dentry->d_lock);
1453 if (atomic_read(&dentry->d_count) > 1) {
1454 spin_unlock(&dentry->d_lock);
1455 spin_unlock(&dcache_lock);
1456 /* Start asynchronous writeout of the inode */
1457 write_inode_now(dentry->d_inode, 0);
1458 error = nfs_sillyrename(dir, dentry);
1462 if (!d_unhashed(dentry)) {
1466 spin_unlock(&dentry->d_lock);
1467 spin_unlock(&dcache_lock);
1468 error = nfs_safe_remove(dentry);
1470 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1471 } else if (need_rehash)
1478 * To create a symbolic link, most file systems instantiate a new inode,
1479 * add a page to it containing the path, then write it out to the disk
1480 * using prepare_write/commit_write.
1482 * Unfortunately the NFS client can't create the in-core inode first
1483 * because it needs a file handle to create an in-core inode (see
1484 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1485 * symlink request has completed on the server.
1487 * So instead we allocate a raw page, copy the symname into it, then do
1488 * the SYMLINK request with the page as the buffer. If it succeeds, we
1489 * now have a new file handle and can instantiate an in-core NFS inode
1490 * and move the raw page into its mapping.
1492 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1494 struct pagevec lru_pvec;
1498 unsigned int pathlen = strlen(symname);
1501 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1502 dir->i_ino, dentry->d_name.name, symname);
1504 if (pathlen > PAGE_SIZE)
1505 return -ENAMETOOLONG;
1507 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1508 attr.ia_valid = ATTR_MODE;
1512 page = alloc_page(GFP_HIGHUSER);
1518 kaddr = kmap_atomic(page, KM_USER0);
1519 memcpy(kaddr, symname, pathlen);
1520 if (pathlen < PAGE_SIZE)
1521 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1522 kunmap_atomic(kaddr, KM_USER0);
1524 nfs_begin_data_update(dir);
1525 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1526 nfs_end_data_update(dir);
1528 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1529 dir->i_sb->s_id, dir->i_ino,
1530 dentry->d_name.name, symname, error);
1538 * No big deal if we can't add this page to the page cache here.
1539 * READLINK will get the missing page from the server if needed.
1541 pagevec_init(&lru_pvec, 0);
1542 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1544 pagevec_add(&lru_pvec, page);
1545 pagevec_lru_add(&lru_pvec);
1546 SetPageUptodate(page);
1556 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1558 struct inode *inode = old_dentry->d_inode;
1561 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1562 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1563 dentry->d_parent->d_name.name, dentry->d_name.name);
1566 nfs_begin_data_update(dir);
1567 nfs_begin_data_update(inode);
1568 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1570 atomic_inc(&inode->i_count);
1571 d_instantiate(dentry, inode);
1573 nfs_end_data_update(inode);
1574 nfs_end_data_update(dir);
1581 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1582 * different file handle for the same inode after a rename (e.g. when
1583 * moving to a different directory). A fail-safe method to do so would
1584 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1585 * rename the old file using the sillyrename stuff. This way, the original
1586 * file in old_dir will go away when the last process iput()s the inode.
1590 * It actually works quite well. One needs to have the possibility for
1591 * at least one ".nfs..." file in each directory the file ever gets
1592 * moved or linked to which happens automagically with the new
1593 * implementation that only depends on the dcache stuff instead of
1594 * using the inode layer
1596 * Unfortunately, things are a little more complicated than indicated
1597 * above. For a cross-directory move, we want to make sure we can get
1598 * rid of the old inode after the operation. This means there must be
1599 * no pending writes (if it's a file), and the use count must be 1.
1600 * If these conditions are met, we can drop the dentries before doing
1603 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1604 struct inode *new_dir, struct dentry *new_dentry)
1606 struct inode *old_inode = old_dentry->d_inode;
1607 struct inode *new_inode = new_dentry->d_inode;
1608 struct dentry *dentry = NULL, *rehash = NULL;
1612 * To prevent any new references to the target during the rename,
1613 * we unhash the dentry and free the inode in advance.
1616 if (!d_unhashed(new_dentry)) {
1618 rehash = new_dentry;
1621 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1622 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1623 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1624 atomic_read(&new_dentry->d_count));
1627 * First check whether the target is busy ... we can't
1628 * safely do _any_ rename if the target is in use.
1630 * For files, make a copy of the dentry and then do a
1631 * silly-rename. If the silly-rename succeeds, the
1632 * copied dentry is hashed and becomes the new target.
1636 if (S_ISDIR(new_inode->i_mode)) {
1638 if (!S_ISDIR(old_inode->i_mode))
1640 } else if (atomic_read(&new_dentry->d_count) > 2) {
1642 /* copy the target dentry's name */
1643 dentry = d_alloc(new_dentry->d_parent,
1644 &new_dentry->d_name);
1648 /* silly-rename the existing target ... */
1649 err = nfs_sillyrename(new_dir, new_dentry);
1651 new_dentry = rehash = dentry;
1653 /* instantiate the replacement target */
1654 d_instantiate(new_dentry, NULL);
1655 } else if (atomic_read(&new_dentry->d_count) > 1)
1656 /* dentry still busy? */
1659 drop_nlink(new_inode);
1663 * ... prune child dentries and writebacks if needed.
1665 if (atomic_read(&old_dentry->d_count) > 1) {
1666 if (S_ISREG(old_inode->i_mode))
1667 nfs_wb_all(old_inode);
1668 shrink_dcache_parent(old_dentry);
1670 nfs_inode_return_delegation(old_inode);
1672 if (new_inode != NULL) {
1673 nfs_inode_return_delegation(new_inode);
1674 d_delete(new_dentry);
1677 nfs_begin_data_update(old_dir);
1678 nfs_begin_data_update(new_dir);
1679 nfs_begin_data_update(old_inode);
1680 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1681 new_dir, &new_dentry->d_name);
1682 nfs_mark_for_revalidate(old_inode);
1683 nfs_end_data_update(old_inode);
1684 nfs_end_data_update(new_dir);
1685 nfs_end_data_update(old_dir);
1690 d_move(old_dentry, new_dentry);
1691 nfs_set_verifier(new_dentry,
1692 nfs_save_change_attribute(new_dir));
1695 /* new dentry created? */
1702 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1703 static LIST_HEAD(nfs_access_lru_list);
1704 static atomic_long_t nfs_access_nr_entries;
1706 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1708 put_rpccred(entry->cred);
1710 smp_mb__before_atomic_dec();
1711 atomic_long_dec(&nfs_access_nr_entries);
1712 smp_mb__after_atomic_dec();
1715 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1718 struct nfs_inode *nfsi;
1719 struct nfs_access_entry *cache;
1722 spin_lock(&nfs_access_lru_lock);
1723 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1724 struct inode *inode;
1726 if (nr_to_scan-- == 0)
1728 inode = igrab(&nfsi->vfs_inode);
1731 spin_lock(&inode->i_lock);
1732 if (list_empty(&nfsi->access_cache_entry_lru))
1733 goto remove_lru_entry;
1734 cache = list_entry(nfsi->access_cache_entry_lru.next,
1735 struct nfs_access_entry, lru);
1736 list_move(&cache->lru, &head);
1737 rb_erase(&cache->rb_node, &nfsi->access_cache);
1738 if (!list_empty(&nfsi->access_cache_entry_lru))
1739 list_move_tail(&nfsi->access_cache_inode_lru,
1740 &nfs_access_lru_list);
1743 list_del_init(&nfsi->access_cache_inode_lru);
1744 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1746 spin_unlock(&inode->i_lock);
1747 spin_unlock(&nfs_access_lru_lock);
1751 spin_unlock(&nfs_access_lru_lock);
1752 while (!list_empty(&head)) {
1753 cache = list_entry(head.next, struct nfs_access_entry, lru);
1754 list_del(&cache->lru);
1755 nfs_access_free_entry(cache);
1757 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1760 static void __nfs_access_zap_cache(struct inode *inode)
1762 struct nfs_inode *nfsi = NFS_I(inode);
1763 struct rb_root *root_node = &nfsi->access_cache;
1764 struct rb_node *n, *dispose = NULL;
1765 struct nfs_access_entry *entry;
1767 /* Unhook entries from the cache */
1768 while ((n = rb_first(root_node)) != NULL) {
1769 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1770 rb_erase(n, root_node);
1771 list_del(&entry->lru);
1772 n->rb_left = dispose;
1775 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1776 spin_unlock(&inode->i_lock);
1778 /* Now kill them all! */
1779 while (dispose != NULL) {
1781 dispose = n->rb_left;
1782 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1786 void nfs_access_zap_cache(struct inode *inode)
1788 /* Remove from global LRU init */
1789 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1790 spin_lock(&nfs_access_lru_lock);
1791 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1792 spin_unlock(&nfs_access_lru_lock);
1795 spin_lock(&inode->i_lock);
1796 /* This will release the spinlock */
1797 __nfs_access_zap_cache(inode);
1800 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1802 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1803 struct nfs_access_entry *entry;
1806 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1808 if (cred < entry->cred)
1810 else if (cred > entry->cred)
1818 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1820 struct nfs_inode *nfsi = NFS_I(inode);
1821 struct nfs_access_entry *cache;
1824 spin_lock(&inode->i_lock);
1825 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1827 cache = nfs_access_search_rbtree(inode, cred);
1830 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1832 res->jiffies = cache->jiffies;
1833 res->cred = cache->cred;
1834 res->mask = cache->mask;
1835 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1838 spin_unlock(&inode->i_lock);
1841 rb_erase(&cache->rb_node, &nfsi->access_cache);
1842 list_del(&cache->lru);
1843 spin_unlock(&inode->i_lock);
1844 nfs_access_free_entry(cache);
1847 /* This will release the spinlock */
1848 __nfs_access_zap_cache(inode);
1852 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1854 struct nfs_inode *nfsi = NFS_I(inode);
1855 struct rb_root *root_node = &nfsi->access_cache;
1856 struct rb_node **p = &root_node->rb_node;
1857 struct rb_node *parent = NULL;
1858 struct nfs_access_entry *entry;
1860 spin_lock(&inode->i_lock);
1861 while (*p != NULL) {
1863 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1865 if (set->cred < entry->cred)
1866 p = &parent->rb_left;
1867 else if (set->cred > entry->cred)
1868 p = &parent->rb_right;
1872 rb_link_node(&set->rb_node, parent, p);
1873 rb_insert_color(&set->rb_node, root_node);
1874 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1875 spin_unlock(&inode->i_lock);
1878 rb_replace_node(parent, &set->rb_node, root_node);
1879 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1880 list_del(&entry->lru);
1881 spin_unlock(&inode->i_lock);
1882 nfs_access_free_entry(entry);
1885 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1887 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1890 RB_CLEAR_NODE(&cache->rb_node);
1891 cache->jiffies = set->jiffies;
1892 cache->cred = get_rpccred(set->cred);
1893 cache->mask = set->mask;
1895 nfs_access_add_rbtree(inode, cache);
1897 /* Update accounting */
1898 smp_mb__before_atomic_inc();
1899 atomic_long_inc(&nfs_access_nr_entries);
1900 smp_mb__after_atomic_inc();
1902 /* Add inode to global LRU list */
1903 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1904 spin_lock(&nfs_access_lru_lock);
1905 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1906 spin_unlock(&nfs_access_lru_lock);
1910 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1912 struct nfs_access_entry cache;
1915 status = nfs_access_get_cached(inode, cred, &cache);
1919 /* Be clever: ask server to check for all possible rights */
1920 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1922 cache.jiffies = jiffies;
1923 status = NFS_PROTO(inode)->access(inode, &cache);
1926 nfs_access_add_cache(inode, &cache);
1928 if ((cache.mask & mask) == mask)
1933 static int nfs_open_permission_mask(int openflags)
1937 if (openflags & FMODE_READ)
1939 if (openflags & FMODE_WRITE)
1941 if (openflags & FMODE_EXEC)
1946 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1948 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1951 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1953 struct rpc_cred *cred;
1956 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1960 /* Is this sys_access() ? */
1961 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1964 switch (inode->i_mode & S_IFMT) {
1968 /* NFSv4 has atomic_open... */
1969 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1971 && (nd->flags & LOOKUP_OPEN))
1976 * Optimize away all write operations, since the server
1977 * will check permissions when we perform the op.
1979 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1986 if (!NFS_PROTO(inode)->access)
1989 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1990 if (!IS_ERR(cred)) {
1991 res = nfs_do_access(inode, cred, mask);
1994 res = PTR_ERR(cred);
1997 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1998 inode->i_sb->s_id, inode->i_ino, mask, res);
2001 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2003 res = generic_permission(inode, mask, NULL);
2010 * version-control: t
2011 * kept-new-versions: 5