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 /* Ensure consistent page alignment of the data.
204 * Note: assumes we have exclusive access to this mapping either
205 * through inode->i_mutex or some other mechanism.
207 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
208 /* Should never happen */
209 nfs_zap_mapping(inode, inode->i_mapping);
216 nfs_zap_caches(inode);
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;
232 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
237 void dir_page_release(nfs_readdir_descriptor_t *desc)
240 page_cache_release(desc->page);
246 * Given a pointer to a buffer that has already been filled by a call
247 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
249 * If the end of the buffer has been reached, return -EAGAIN, if not,
250 * return the offset within the buffer of the next entry to be
254 int find_dirent(nfs_readdir_descriptor_t *desc)
256 struct nfs_entry *entry = desc->entry;
260 while((status = dir_decode(desc)) == 0) {
261 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
262 __FUNCTION__, (unsigned long long)entry->cookie);
263 if (entry->prev_cookie == *desc->dir_cookie)
265 if (loop_count++ > 200) {
274 * Given a pointer to a buffer that has already been filled by a call
275 * to readdir, find the entry at offset 'desc->file->f_pos'.
277 * If the end of the buffer has been reached, return -EAGAIN, if not,
278 * return the offset within the buffer of the next entry to be
282 int find_dirent_index(nfs_readdir_descriptor_t *desc)
284 struct nfs_entry *entry = desc->entry;
289 status = dir_decode(desc);
293 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
294 (unsigned long long)entry->cookie, desc->current_index);
296 if (desc->file->f_pos == desc->current_index) {
297 *desc->dir_cookie = entry->cookie;
300 desc->current_index++;
301 if (loop_count++ > 200) {
310 * Find the given page, and call find_dirent() or find_dirent_index in
311 * order to try to return the next entry.
314 int find_dirent_page(nfs_readdir_descriptor_t *desc)
316 struct inode *inode = desc->file->f_path.dentry->d_inode;
320 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
321 __FUNCTION__, desc->page_index,
322 (long long) *desc->dir_cookie);
324 /* If we find the page in the page_cache, we cannot be sure
325 * how fresh the data is, so we will ignore readdir_plus attributes.
327 desc->timestamp_valid = 0;
328 page = read_cache_page(inode->i_mapping, desc->page_index,
329 (filler_t *)nfs_readdir_filler, desc);
331 status = PTR_ERR(page);
335 /* NOTE: Someone else may have changed the READDIRPLUS flag */
337 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
338 if (*desc->dir_cookie != 0)
339 status = find_dirent(desc);
341 status = find_dirent_index(desc);
343 dir_page_release(desc);
345 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
350 * Recurse through the page cache pages, and return a
351 * filled nfs_entry structure of the next directory entry if possible.
353 * The target for the search is '*desc->dir_cookie' if non-0,
354 * 'desc->file->f_pos' otherwise
357 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
362 /* Always search-by-index from the beginning of the cache */
363 if (*desc->dir_cookie == 0) {
364 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
365 (long long)desc->file->f_pos);
366 desc->page_index = 0;
367 desc->entry->cookie = desc->entry->prev_cookie = 0;
368 desc->entry->eof = 0;
369 desc->current_index = 0;
371 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
372 (unsigned long long)*desc->dir_cookie);
375 res = find_dirent_page(desc);
378 /* Align to beginning of next page */
380 if (loop_count++ > 200) {
386 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
390 static inline unsigned int dt_type(struct inode *inode)
392 return (inode->i_mode >> 12) & 15;
395 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
398 * Once we've found the start of the dirent within a page: fill 'er up...
401 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
404 struct file *file = desc->file;
405 struct nfs_entry *entry = desc->entry;
406 struct dentry *dentry = NULL;
411 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
412 (unsigned long long)entry->cookie);
415 unsigned d_type = DT_UNKNOWN;
416 /* Note: entry->prev_cookie contains the cookie for
417 * retrieving the current dirent on the server */
420 /* Get a dentry if we have one */
423 dentry = nfs_readdir_lookup(desc);
425 /* Use readdirplus info */
426 if (dentry != NULL && dentry->d_inode != NULL) {
427 d_type = dt_type(dentry->d_inode);
428 fileid = NFS_FILEID(dentry->d_inode);
431 res = filldir(dirent, entry->name, entry->len,
432 file->f_pos, fileid, d_type);
436 *desc->dir_cookie = entry->cookie;
437 if (dir_decode(desc) != 0) {
441 if (loop_count++ > 200) {
446 dir_page_release(desc);
449 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
450 (unsigned long long)*desc->dir_cookie, res);
455 * If we cannot find a cookie in our cache, we suspect that this is
456 * because it points to a deleted file, so we ask the server to return
457 * whatever it thinks is the next entry. We then feed this to filldir.
458 * If all goes well, we should then be able to find our way round the
459 * cache on the next call to readdir_search_pagecache();
461 * NOTE: we cannot add the anonymous page to the pagecache because
462 * the data it contains might not be page aligned. Besides,
463 * we should already have a complete representation of the
464 * directory in the page cache by the time we get here.
467 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
470 struct file *file = desc->file;
471 struct inode *inode = file->f_path.dentry->d_inode;
472 struct rpc_cred *cred = nfs_file_cred(file);
473 struct page *page = NULL;
475 unsigned long timestamp;
477 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
478 (unsigned long long)*desc->dir_cookie);
480 page = alloc_page(GFP_HIGHUSER);
486 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
488 NFS_SERVER(inode)->dtsize,
491 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
492 if (desc->error >= 0) {
493 desc->timestamp = timestamp;
494 desc->timestamp_valid = 1;
495 if ((status = dir_decode(desc)) == 0)
496 desc->entry->prev_cookie = *desc->dir_cookie;
502 status = nfs_do_filldir(desc, dirent, filldir);
504 /* Reset read descriptor so it searches the page cache from
505 * the start upon the next call to readdir_search_pagecache() */
506 desc->page_index = 0;
507 desc->entry->cookie = desc->entry->prev_cookie = 0;
508 desc->entry->eof = 0;
510 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
511 __FUNCTION__, status);
514 dir_page_release(desc);
518 /* The file offset position represents the dirent entry number. A
519 last cookie cache takes care of the common case of reading the
522 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
524 struct dentry *dentry = filp->f_path.dentry;
525 struct inode *inode = dentry->d_inode;
526 nfs_readdir_descriptor_t my_desc,
528 struct nfs_entry my_entry;
530 struct nfs_fattr fattr;
533 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
534 dentry->d_parent->d_name.name, dentry->d_name.name,
535 (long long)filp->f_pos);
536 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
540 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
547 * filp->f_pos points to the dirent entry number.
548 * *desc->dir_cookie has the cookie for the next entry. We have
549 * to either find the entry with the appropriate number or
550 * revalidate the cookie.
552 memset(desc, 0, sizeof(*desc));
555 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
556 desc->decode = NFS_PROTO(inode)->decode_dirent;
557 desc->plus = NFS_USE_READDIRPLUS(inode);
559 my_entry.cookie = my_entry.prev_cookie = 0;
562 my_entry.fattr = &fattr;
563 nfs_fattr_init(&fattr);
564 desc->entry = &my_entry;
566 while(!desc->entry->eof) {
567 res = readdir_search_pagecache(desc);
569 if (res == -EBADCOOKIE) {
570 /* This means either end of directory */
571 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
572 /* Or that the server has 'lost' a cookie */
573 res = uncached_readdir(desc, dirent, filldir);
580 if (res == -ETOOSMALL && desc->plus) {
581 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
582 nfs_zap_caches(inode);
584 desc->entry->eof = 0;
590 res = nfs_do_filldir(desc, dirent, filldir);
599 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
600 dentry->d_parent->d_name.name, dentry->d_name.name,
605 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
607 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
610 offset += filp->f_pos;
618 if (offset != filp->f_pos) {
619 filp->f_pos = offset;
620 nfs_file_open_context(filp)->dir_cookie = 0;
623 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
628 * All directory operations under NFS are synchronous, so fsync()
629 * is a dummy operation.
631 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
633 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
634 dentry->d_parent->d_name.name, dentry->d_name.name,
641 * A check for whether or not the parent directory has changed.
642 * In the case it has, we assume that the dentries are untrustworthy
643 * and may need to be looked up again.
645 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
649 if (dentry->d_time == NFS_I(dir)->cache_change_attribute)
654 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
656 dentry->d_time = verf;
660 * Return the intent data that applies to this particular path component
662 * Note that the current set of intents only apply to the very last
663 * component of the path.
664 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
666 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
668 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
670 return nd->flags & mask;
674 * Inode and filehandle revalidation for lookups.
676 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
677 * or if the intent information indicates that we're about to open this
678 * particular file and the "nocto" mount flag is not set.
682 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
684 struct nfs_server *server = NFS_SERVER(inode);
687 /* VFS wants an on-the-wire revalidation */
688 if (nd->flags & LOOKUP_REVAL)
690 /* This is an open(2) */
691 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
692 !(server->flags & NFS_MOUNT_NOCTO) &&
693 (S_ISREG(inode->i_mode) ||
694 S_ISDIR(inode->i_mode)))
697 return nfs_revalidate_inode(server, inode);
699 return __nfs_revalidate_inode(server, inode);
703 * We judge how long we want to trust negative
704 * dentries by looking at the parent inode mtime.
706 * If parent mtime has changed, we revalidate, else we wait for a
707 * period corresponding to the parent's attribute cache timeout value.
710 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
711 struct nameidata *nd)
713 /* Don't revalidate a negative dentry if we're creating a new file */
714 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
716 return !nfs_check_verifier(dir, dentry);
720 * This is called every time the dcache has a lookup hit,
721 * and we should check whether we can really trust that
724 * NOTE! The hit can be a negative hit too, don't assume
727 * If the parent directory is seen to have changed, we throw out the
728 * cached dentry and do a new lookup.
730 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
734 struct dentry *parent;
736 struct nfs_fh fhandle;
737 struct nfs_fattr fattr;
738 unsigned long verifier;
740 parent = dget_parent(dentry);
742 dir = parent->d_inode;
743 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
744 inode = dentry->d_inode;
746 /* Revalidate parent directory attribute cache */
747 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
751 if (nfs_neg_need_reval(dir, dentry, nd))
756 if (is_bad_inode(inode)) {
757 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
758 __FUNCTION__, dentry->d_parent->d_name.name,
759 dentry->d_name.name);
763 /* Force a full look up iff the parent directory has changed */
764 if (nfs_check_verifier(dir, dentry)) {
765 if (nfs_lookup_verify_inode(inode, nd))
770 if (NFS_STALE(inode))
773 verifier = nfs_save_change_attribute(dir);
774 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
777 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
779 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
782 nfs_set_verifier(dentry, verifier);
786 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
787 __FUNCTION__, dentry->d_parent->d_name.name,
788 dentry->d_name.name);
794 if (inode && S_ISDIR(inode->i_mode)) {
795 /* Purge readdir caches. */
796 nfs_zap_caches(inode);
797 /* If we have submounts, don't unhash ! */
798 if (have_submounts(dentry))
800 shrink_dcache_parent(dentry);
805 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
806 __FUNCTION__, dentry->d_parent->d_name.name,
807 dentry->d_name.name);
812 * This is called from dput() when d_count is going to 0.
814 static int nfs_dentry_delete(struct dentry *dentry)
816 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
817 dentry->d_parent->d_name.name, dentry->d_name.name,
820 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
821 /* Unhash it, so that ->d_iput() would be called */
824 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
825 /* Unhash it, so that ancestors of killed async unlink
826 * files will be cleaned up during umount */
834 * Called when the dentry loses inode.
835 * We use it to clean up silly-renamed files.
837 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
839 nfs_inode_return_delegation(inode);
840 if (S_ISDIR(inode->i_mode))
841 /* drop any readdir cache as it could easily be old */
842 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
844 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
847 nfs_complete_unlink(dentry, inode);
853 struct dentry_operations nfs_dentry_operations = {
854 .d_revalidate = nfs_lookup_revalidate,
855 .d_delete = nfs_dentry_delete,
856 .d_iput = nfs_dentry_iput,
860 * Use intent information to check whether or not we're going to do
861 * an O_EXCL create using this path component.
864 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
866 if (NFS_PROTO(dir)->version == 2)
868 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
870 return (nd->intent.open.flags & O_EXCL) != 0;
873 static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
875 struct nfs_server *server = NFS_SERVER(dir);
877 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
878 /* Revalidate fsid using the parent directory */
879 return __nfs_revalidate_inode(server, dir);
883 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
886 struct inode *inode = NULL;
888 struct nfs_fh fhandle;
889 struct nfs_fattr fattr;
891 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
892 dentry->d_parent->d_name.name, dentry->d_name.name);
893 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
895 res = ERR_PTR(-ENAMETOOLONG);
896 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
899 res = ERR_PTR(-ENOMEM);
900 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
905 * If we're doing an exclusive create, optimize away the lookup
906 * but don't hash the dentry.
908 if (nfs_is_exclusive_create(dir, nd)) {
909 d_instantiate(dentry, NULL);
914 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
915 if (error == -ENOENT)
918 res = ERR_PTR(error);
921 error = nfs_reval_fsid(dir, &fattr);
923 res = ERR_PTR(error);
926 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
927 res = (struct dentry *)inode;
932 res = d_materialise_unique(dentry, inode);
934 struct dentry *parent;
937 /* Was a directory renamed! */
938 parent = dget_parent(res);
939 if (!IS_ROOT(parent))
940 nfs_mark_for_revalidate(parent->d_inode);
944 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
952 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
954 struct dentry_operations nfs4_dentry_operations = {
955 .d_revalidate = nfs_open_revalidate,
956 .d_delete = nfs_dentry_delete,
957 .d_iput = nfs_dentry_iput,
961 * Use intent information to determine whether we need to substitute
962 * the NFSv4-style stateful OPEN for the LOOKUP call
964 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
966 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
968 /* NFS does not (yet) have a stateful open for directories */
969 if (nd->flags & LOOKUP_DIRECTORY)
971 /* Are we trying to write to a read only partition? */
972 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
977 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
979 struct dentry *res = NULL;
982 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
983 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
985 /* Check that we are indeed trying to open this file */
986 if (!is_atomic_open(dir, nd))
989 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
990 res = ERR_PTR(-ENAMETOOLONG);
993 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
995 /* Let vfs_create() deal with O_EXCL */
996 if (nd->intent.open.flags & O_EXCL) {
1001 /* Open the file on the server */
1003 /* Revalidate parent directory attribute cache */
1004 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1006 res = ERR_PTR(error);
1011 if (nd->intent.open.flags & O_CREAT) {
1012 nfs_begin_data_update(dir);
1013 res = nfs4_atomic_open(dir, dentry, nd);
1014 nfs_end_data_update(dir);
1016 res = nfs4_atomic_open(dir, dentry, nd);
1019 error = PTR_ERR(res);
1021 /* Make a negative dentry */
1025 /* This turned out not to be a regular file */
1030 if (!(nd->intent.open.flags & O_NOFOLLOW))
1036 } else if (res != NULL)
1038 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1042 return nfs_lookup(dir, dentry, nd);
1045 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1047 struct dentry *parent = NULL;
1048 struct inode *inode = dentry->d_inode;
1050 unsigned long verifier;
1051 int openflags, ret = 0;
1053 parent = dget_parent(dentry);
1054 dir = parent->d_inode;
1055 if (!is_atomic_open(dir, nd))
1057 /* We can't create new files in nfs_open_revalidate(), so we
1058 * optimize away revalidation of negative dentries.
1062 /* NFS only supports OPEN on regular files */
1063 if (!S_ISREG(inode->i_mode))
1065 openflags = nd->intent.open.flags;
1066 /* We cannot do exclusive creation on a positive dentry */
1067 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1069 /* We can't create new files, or truncate existing ones here */
1070 openflags &= ~(O_CREAT|O_TRUNC);
1073 * Note: we're not holding inode->i_mutex and so may be racing with
1074 * operations that change the directory. We therefore save the
1075 * change attribute *before* we do the RPC call.
1078 verifier = nfs_save_change_attribute(dir);
1079 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1081 nfs_set_verifier(dentry, verifier);
1090 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
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 inode *inode;
1178 int error = -EACCES;
1180 /* We may have been initialized further down */
1181 if (dentry->d_inode)
1183 if (fhandle->size == 0) {
1184 struct inode *dir = dentry->d_parent->d_inode;
1185 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1189 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1190 struct nfs_server *server = NFS_SB(dentry->d_sb);
1191 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1195 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1196 error = PTR_ERR(inode);
1199 d_instantiate(dentry, inode);
1200 if (d_unhashed(dentry))
1206 * Following a failed create operation, we drop the dentry rather
1207 * than retain a negative dentry. This avoids a problem in the event
1208 * that the operation succeeded on the server, but an error in the
1209 * reply path made it appear to have failed.
1211 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1212 struct nameidata *nd)
1218 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1219 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1221 attr.ia_mode = mode;
1222 attr.ia_valid = ATTR_MODE;
1224 if ((nd->flags & LOOKUP_CREATE) != 0)
1225 open_flags = nd->intent.open.flags;
1228 nfs_begin_data_update(dir);
1229 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1230 nfs_end_data_update(dir);
1233 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1243 * See comments for nfs_proc_create regarding failed operations.
1246 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1251 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1252 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1254 if (!new_valid_dev(rdev))
1257 attr.ia_mode = mode;
1258 attr.ia_valid = ATTR_MODE;
1261 nfs_begin_data_update(dir);
1262 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1263 nfs_end_data_update(dir);
1266 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1276 * See comments for nfs_proc_create regarding failed operations.
1278 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1283 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1284 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1286 attr.ia_valid = ATTR_MODE;
1287 attr.ia_mode = mode | S_IFDIR;
1290 nfs_begin_data_update(dir);
1291 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1292 nfs_end_data_update(dir);
1295 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1304 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1308 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1309 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1312 nfs_begin_data_update(dir);
1313 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1314 /* Ensure the VFS deletes this inode */
1315 if (error == 0 && dentry->d_inode != NULL)
1316 clear_nlink(dentry->d_inode);
1317 nfs_end_data_update(dir);
1323 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1325 static unsigned int sillycounter;
1326 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1327 const int countersize = sizeof(sillycounter)*2;
1328 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1331 struct dentry *sdentry;
1334 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1335 dentry->d_parent->d_name.name, dentry->d_name.name,
1336 atomic_read(&dentry->d_count));
1337 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1340 * We don't allow a dentry to be silly-renamed twice.
1343 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1346 sprintf(silly, ".nfs%*.*Lx",
1347 fileidsize, fileidsize,
1348 (unsigned long long)NFS_FILEID(dentry->d_inode));
1350 /* Return delegation in anticipation of the rename */
1351 nfs_inode_return_delegation(dentry->d_inode);
1355 char *suffix = silly + slen - countersize;
1359 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1361 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1362 dentry->d_name.name, silly);
1364 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1366 * N.B. Better to return EBUSY here ... it could be
1367 * dangerous to delete the file while it's in use.
1369 if (IS_ERR(sdentry))
1371 } while(sdentry->d_inode != NULL); /* need negative lookup */
1373 qsilly.name = silly;
1374 qsilly.len = strlen(silly);
1375 nfs_begin_data_update(dir);
1376 if (dentry->d_inode) {
1377 nfs_begin_data_update(dentry->d_inode);
1378 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1380 nfs_mark_for_revalidate(dentry->d_inode);
1381 nfs_end_data_update(dentry->d_inode);
1383 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1385 nfs_end_data_update(dir);
1387 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1388 d_move(dentry, sdentry);
1389 error = nfs_async_unlink(dir, dentry);
1390 /* If we return 0 we don't unlink */
1398 * Remove a file after making sure there are no pending writes,
1399 * and after checking that the file has only one user.
1401 * We invalidate the attribute cache and free the inode prior to the operation
1402 * to avoid possible races if the server reuses the inode.
1404 static int nfs_safe_remove(struct dentry *dentry)
1406 struct inode *dir = dentry->d_parent->d_inode;
1407 struct inode *inode = dentry->d_inode;
1410 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1411 dentry->d_parent->d_name.name, dentry->d_name.name);
1413 /* If the dentry was sillyrenamed, we simply call d_delete() */
1414 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1419 nfs_begin_data_update(dir);
1420 if (inode != NULL) {
1421 nfs_inode_return_delegation(inode);
1422 nfs_begin_data_update(inode);
1423 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1424 /* The VFS may want to delete this inode */
1427 nfs_mark_for_revalidate(inode);
1428 nfs_end_data_update(inode);
1430 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1431 nfs_end_data_update(dir);
1436 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1437 * belongs to an active ".nfs..." file and we return -EBUSY.
1439 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1441 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1444 int need_rehash = 0;
1446 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1447 dir->i_ino, dentry->d_name.name);
1450 spin_lock(&dcache_lock);
1451 spin_lock(&dentry->d_lock);
1452 if (atomic_read(&dentry->d_count) > 1) {
1453 spin_unlock(&dentry->d_lock);
1454 spin_unlock(&dcache_lock);
1455 /* Start asynchronous writeout of the inode */
1456 write_inode_now(dentry->d_inode, 0);
1457 error = nfs_sillyrename(dir, dentry);
1461 if (!d_unhashed(dentry)) {
1465 spin_unlock(&dentry->d_lock);
1466 spin_unlock(&dcache_lock);
1467 error = nfs_safe_remove(dentry);
1469 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1470 } else if (need_rehash)
1477 * To create a symbolic link, most file systems instantiate a new inode,
1478 * add a page to it containing the path, then write it out to the disk
1479 * using prepare_write/commit_write.
1481 * Unfortunately the NFS client can't create the in-core inode first
1482 * because it needs a file handle to create an in-core inode (see
1483 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1484 * symlink request has completed on the server.
1486 * So instead we allocate a raw page, copy the symname into it, then do
1487 * the SYMLINK request with the page as the buffer. If it succeeds, we
1488 * now have a new file handle and can instantiate an in-core NFS inode
1489 * and move the raw page into its mapping.
1491 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1493 struct pagevec lru_pvec;
1497 unsigned int pathlen = strlen(symname);
1500 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1501 dir->i_ino, dentry->d_name.name, symname);
1503 if (pathlen > PAGE_SIZE)
1504 return -ENAMETOOLONG;
1506 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1507 attr.ia_valid = ATTR_MODE;
1511 page = alloc_page(GFP_HIGHUSER);
1517 kaddr = kmap_atomic(page, KM_USER0);
1518 memcpy(kaddr, symname, pathlen);
1519 if (pathlen < PAGE_SIZE)
1520 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1521 kunmap_atomic(kaddr, KM_USER0);
1523 nfs_begin_data_update(dir);
1524 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1525 nfs_end_data_update(dir);
1527 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1528 dir->i_sb->s_id, dir->i_ino,
1529 dentry->d_name.name, symname, error);
1537 * No big deal if we can't add this page to the page cache here.
1538 * READLINK will get the missing page from the server if needed.
1540 pagevec_init(&lru_pvec, 0);
1541 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1543 pagevec_add(&lru_pvec, page);
1544 pagevec_lru_add(&lru_pvec);
1545 SetPageUptodate(page);
1555 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1557 struct inode *inode = old_dentry->d_inode;
1560 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1561 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1562 dentry->d_parent->d_name.name, dentry->d_name.name);
1565 nfs_begin_data_update(dir);
1566 nfs_begin_data_update(inode);
1567 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1569 atomic_inc(&inode->i_count);
1570 d_instantiate(dentry, inode);
1572 nfs_end_data_update(inode);
1573 nfs_end_data_update(dir);
1580 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1581 * different file handle for the same inode after a rename (e.g. when
1582 * moving to a different directory). A fail-safe method to do so would
1583 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1584 * rename the old file using the sillyrename stuff. This way, the original
1585 * file in old_dir will go away when the last process iput()s the inode.
1589 * It actually works quite well. One needs to have the possibility for
1590 * at least one ".nfs..." file in each directory the file ever gets
1591 * moved or linked to which happens automagically with the new
1592 * implementation that only depends on the dcache stuff instead of
1593 * using the inode layer
1595 * Unfortunately, things are a little more complicated than indicated
1596 * above. For a cross-directory move, we want to make sure we can get
1597 * rid of the old inode after the operation. This means there must be
1598 * no pending writes (if it's a file), and the use count must be 1.
1599 * If these conditions are met, we can drop the dentries before doing
1602 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1603 struct inode *new_dir, struct dentry *new_dentry)
1605 struct inode *old_inode = old_dentry->d_inode;
1606 struct inode *new_inode = new_dentry->d_inode;
1607 struct dentry *dentry = NULL, *rehash = NULL;
1611 * To prevent any new references to the target during the rename,
1612 * we unhash the dentry and free the inode in advance.
1615 if (!d_unhashed(new_dentry)) {
1617 rehash = new_dentry;
1620 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1621 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1622 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1623 atomic_read(&new_dentry->d_count));
1626 * First check whether the target is busy ... we can't
1627 * safely do _any_ rename if the target is in use.
1629 * For files, make a copy of the dentry and then do a
1630 * silly-rename. If the silly-rename succeeds, the
1631 * copied dentry is hashed and becomes the new target.
1635 if (S_ISDIR(new_inode->i_mode)) {
1637 if (!S_ISDIR(old_inode->i_mode))
1639 } else if (atomic_read(&new_dentry->d_count) > 2) {
1641 /* copy the target dentry's name */
1642 dentry = d_alloc(new_dentry->d_parent,
1643 &new_dentry->d_name);
1647 /* silly-rename the existing target ... */
1648 err = nfs_sillyrename(new_dir, new_dentry);
1650 new_dentry = rehash = dentry;
1652 /* instantiate the replacement target */
1653 d_instantiate(new_dentry, NULL);
1654 } else if (atomic_read(&new_dentry->d_count) > 1)
1655 /* dentry still busy? */
1658 drop_nlink(new_inode);
1662 * ... prune child dentries and writebacks if needed.
1664 if (atomic_read(&old_dentry->d_count) > 1) {
1665 if (S_ISREG(old_inode->i_mode))
1666 nfs_wb_all(old_inode);
1667 shrink_dcache_parent(old_dentry);
1669 nfs_inode_return_delegation(old_inode);
1671 if (new_inode != NULL) {
1672 nfs_inode_return_delegation(new_inode);
1673 d_delete(new_dentry);
1676 nfs_begin_data_update(old_dir);
1677 nfs_begin_data_update(new_dir);
1678 nfs_begin_data_update(old_inode);
1679 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1680 new_dir, &new_dentry->d_name);
1681 nfs_mark_for_revalidate(old_inode);
1682 nfs_end_data_update(old_inode);
1683 nfs_end_data_update(new_dir);
1684 nfs_end_data_update(old_dir);
1689 d_move(old_dentry, new_dentry);
1690 nfs_set_verifier(new_dentry,
1691 nfs_save_change_attribute(new_dir));
1694 /* new dentry created? */
1701 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1702 static LIST_HEAD(nfs_access_lru_list);
1703 static atomic_long_t nfs_access_nr_entries;
1705 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1707 put_rpccred(entry->cred);
1709 smp_mb__before_atomic_dec();
1710 atomic_long_dec(&nfs_access_nr_entries);
1711 smp_mb__after_atomic_dec();
1714 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1717 struct nfs_inode *nfsi;
1718 struct nfs_access_entry *cache;
1721 spin_lock(&nfs_access_lru_lock);
1722 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1723 struct inode *inode;
1725 if (nr_to_scan-- == 0)
1727 inode = igrab(&nfsi->vfs_inode);
1730 spin_lock(&inode->i_lock);
1731 if (list_empty(&nfsi->access_cache_entry_lru))
1732 goto remove_lru_entry;
1733 cache = list_entry(nfsi->access_cache_entry_lru.next,
1734 struct nfs_access_entry, lru);
1735 list_move(&cache->lru, &head);
1736 rb_erase(&cache->rb_node, &nfsi->access_cache);
1737 if (!list_empty(&nfsi->access_cache_entry_lru))
1738 list_move_tail(&nfsi->access_cache_inode_lru,
1739 &nfs_access_lru_list);
1742 list_del_init(&nfsi->access_cache_inode_lru);
1743 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1745 spin_unlock(&inode->i_lock);
1746 spin_unlock(&nfs_access_lru_lock);
1750 spin_unlock(&nfs_access_lru_lock);
1751 while (!list_empty(&head)) {
1752 cache = list_entry(head.next, struct nfs_access_entry, lru);
1753 list_del(&cache->lru);
1754 nfs_access_free_entry(cache);
1756 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1759 static void __nfs_access_zap_cache(struct inode *inode)
1761 struct nfs_inode *nfsi = NFS_I(inode);
1762 struct rb_root *root_node = &nfsi->access_cache;
1763 struct rb_node *n, *dispose = NULL;
1764 struct nfs_access_entry *entry;
1766 /* Unhook entries from the cache */
1767 while ((n = rb_first(root_node)) != NULL) {
1768 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1769 rb_erase(n, root_node);
1770 list_del(&entry->lru);
1771 n->rb_left = dispose;
1774 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1775 spin_unlock(&inode->i_lock);
1777 /* Now kill them all! */
1778 while (dispose != NULL) {
1780 dispose = n->rb_left;
1781 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1785 void nfs_access_zap_cache(struct inode *inode)
1787 /* Remove from global LRU init */
1788 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1789 spin_lock(&nfs_access_lru_lock);
1790 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1791 spin_unlock(&nfs_access_lru_lock);
1794 spin_lock(&inode->i_lock);
1795 /* This will release the spinlock */
1796 __nfs_access_zap_cache(inode);
1799 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1801 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1802 struct nfs_access_entry *entry;
1805 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1807 if (cred < entry->cred)
1809 else if (cred > entry->cred)
1817 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1819 struct nfs_inode *nfsi = NFS_I(inode);
1820 struct nfs_access_entry *cache;
1823 spin_lock(&inode->i_lock);
1824 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1826 cache = nfs_access_search_rbtree(inode, cred);
1829 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1831 res->jiffies = cache->jiffies;
1832 res->cred = cache->cred;
1833 res->mask = cache->mask;
1834 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1837 spin_unlock(&inode->i_lock);
1840 rb_erase(&cache->rb_node, &nfsi->access_cache);
1841 list_del(&cache->lru);
1842 spin_unlock(&inode->i_lock);
1843 nfs_access_free_entry(cache);
1846 /* This will release the spinlock */
1847 __nfs_access_zap_cache(inode);
1851 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1853 struct nfs_inode *nfsi = NFS_I(inode);
1854 struct rb_root *root_node = &nfsi->access_cache;
1855 struct rb_node **p = &root_node->rb_node;
1856 struct rb_node *parent = NULL;
1857 struct nfs_access_entry *entry;
1859 spin_lock(&inode->i_lock);
1860 while (*p != NULL) {
1862 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1864 if (set->cred < entry->cred)
1865 p = &parent->rb_left;
1866 else if (set->cred > entry->cred)
1867 p = &parent->rb_right;
1871 rb_link_node(&set->rb_node, parent, p);
1872 rb_insert_color(&set->rb_node, root_node);
1873 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1874 spin_unlock(&inode->i_lock);
1877 rb_replace_node(parent, &set->rb_node, root_node);
1878 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1879 list_del(&entry->lru);
1880 spin_unlock(&inode->i_lock);
1881 nfs_access_free_entry(entry);
1884 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1886 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1889 RB_CLEAR_NODE(&cache->rb_node);
1890 cache->jiffies = set->jiffies;
1891 cache->cred = get_rpccred(set->cred);
1892 cache->mask = set->mask;
1894 nfs_access_add_rbtree(inode, cache);
1896 /* Update accounting */
1897 smp_mb__before_atomic_inc();
1898 atomic_long_inc(&nfs_access_nr_entries);
1899 smp_mb__after_atomic_inc();
1901 /* Add inode to global LRU list */
1902 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1903 spin_lock(&nfs_access_lru_lock);
1904 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1905 spin_unlock(&nfs_access_lru_lock);
1909 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1911 struct nfs_access_entry cache;
1914 status = nfs_access_get_cached(inode, cred, &cache);
1918 /* Be clever: ask server to check for all possible rights */
1919 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1921 cache.jiffies = jiffies;
1922 status = NFS_PROTO(inode)->access(inode, &cache);
1925 nfs_access_add_cache(inode, &cache);
1927 if ((cache.mask & mask) == mask)
1932 static int nfs_open_permission_mask(int openflags)
1936 if (openflags & FMODE_READ)
1938 if (openflags & FMODE_WRITE)
1940 if (openflags & FMODE_EXEC)
1945 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1947 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1950 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1952 struct rpc_cred *cred;
1955 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1959 /* Is this sys_access() ? */
1960 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1963 switch (inode->i_mode & S_IFMT) {
1967 /* NFSv4 has atomic_open... */
1968 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1970 && (nd->flags & LOOKUP_OPEN))
1975 * Optimize away all write operations, since the server
1976 * will check permissions when we perform the op.
1978 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1985 if (!NFS_PROTO(inode)->access)
1988 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1989 if (!IS_ERR(cred)) {
1990 res = nfs_do_access(inode, cred, mask);
1993 res = PTR_ERR(cred);
1996 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1997 inode->i_sb->s_id, inode->i_ino, mask, res);
2000 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2002 res = generic_permission(inode, mask, NULL);
2009 * version-control: t
2010 * kept-new-versions: 5