4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
39 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode *, struct file *);
46 static int nfs_readdir(struct file *, void *, filldir_t);
47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49 static int nfs_mkdir(struct inode *, struct dentry *, int);
50 static int nfs_rmdir(struct inode *, struct dentry *);
51 static int nfs_unlink(struct inode *, struct dentry *);
52 static int nfs_symlink(struct inode *, struct dentry *, const char *);
53 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55 static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57 static int nfs_fsync_dir(struct file *, struct dentry *, int);
58 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 const struct file_operations nfs_dir_operations = {
61 .llseek = nfs_llseek_dir,
62 .read = generic_read_dir,
63 .readdir = nfs_readdir,
65 .release = nfs_release,
66 .fsync = nfs_fsync_dir,
69 const struct inode_operations nfs_dir_inode_operations = {
74 .symlink = nfs_symlink,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
85 const struct inode_operations nfs3_dir_inode_operations = {
90 .symlink = nfs_symlink,
95 .permission = nfs_permission,
96 .getattr = nfs_getattr,
97 .setattr = nfs_setattr,
98 .listxattr = nfs3_listxattr,
99 .getxattr = nfs3_getxattr,
100 .setxattr = nfs3_setxattr,
101 .removexattr = nfs3_removexattr,
103 #endif /* CONFIG_NFS_V3 */
107 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
108 const struct inode_operations nfs4_dir_inode_operations = {
109 .create = nfs_create,
110 .lookup = nfs_atomic_lookup,
112 .unlink = nfs_unlink,
113 .symlink = nfs_symlink,
117 .rename = nfs_rename,
118 .permission = nfs_permission,
119 .getattr = nfs_getattr,
120 .setattr = nfs_setattr,
121 .getxattr = nfs4_getxattr,
122 .setxattr = nfs4_setxattr,
123 .listxattr = nfs4_listxattr,
126 #endif /* CONFIG_NFS_V4 */
132 nfs_opendir(struct inode *inode, struct file *filp)
136 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
137 filp->f_path.dentry->d_parent->d_name.name,
138 filp->f_path.dentry->d_name.name);
140 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
143 /* Call generic open code in order to cache credentials */
144 res = nfs_open(inode, filp);
149 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
153 unsigned long page_index;
156 loff_t current_index;
157 struct nfs_entry *entry;
158 decode_dirent_t decode;
160 unsigned long timestamp;
162 } nfs_readdir_descriptor_t;
164 /* Now we cache directories properly, by stuffing the dirent
165 * data directly in the page cache.
167 * Inode invalidation due to refresh etc. takes care of
168 * _everything_, no sloppy entry flushing logic, no extraneous
169 * copying, network direct to page cache, the way it was meant
172 * NOTE: Dirent information verification is done always by the
173 * page-in of the RPC reply, nowhere else, this simplies
174 * things substantially.
177 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
179 struct file *file = desc->file;
180 struct inode *inode = file->f_path.dentry->d_inode;
181 struct rpc_cred *cred = nfs_file_cred(file);
182 unsigned long timestamp;
185 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
186 __func__, (long long)desc->entry->cookie,
191 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
192 NFS_SERVER(inode)->dtsize, desc->plus);
194 /* We requested READDIRPLUS, but the server doesn't grok it */
195 if (error == -ENOTSUPP && desc->plus) {
196 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
197 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
203 desc->timestamp = timestamp;
204 desc->timestamp_valid = 1;
205 SetPageUptodate(page);
206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
208 * through inode->i_mutex or some other mechanism.
210 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
222 int dir_decode(nfs_readdir_descriptor_t *desc)
224 __be32 *p = desc->ptr;
225 p = desc->decode(p, desc->entry, desc->plus);
229 if (desc->timestamp_valid)
230 desc->entry->fattr->time_start = desc->timestamp;
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 __func__, (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 __func__, 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", __func__, 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", __func__, 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, nfs_compat_user_ino64(fileid),
437 *desc->dir_cookie = entry->cookie;
438 if (dir_decode(desc) != 0) {
442 if (loop_count++ > 200) {
447 dir_page_release(desc);
450 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
451 (unsigned long long)*desc->dir_cookie, res);
456 * If we cannot find a cookie in our cache, we suspect that this is
457 * because it points to a deleted file, so we ask the server to return
458 * whatever it thinks is the next entry. We then feed this to filldir.
459 * If all goes well, we should then be able to find our way round the
460 * cache on the next call to readdir_search_pagecache();
462 * NOTE: we cannot add the anonymous page to the pagecache because
463 * the data it contains might not be page aligned. Besides,
464 * we should already have a complete representation of the
465 * directory in the page cache by the time we get here.
468 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
471 struct file *file = desc->file;
472 struct inode *inode = file->f_path.dentry->d_inode;
473 struct rpc_cred *cred = nfs_file_cred(file);
474 struct page *page = NULL;
476 unsigned long timestamp;
478 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
479 (unsigned long long)*desc->dir_cookie);
481 page = alloc_page(GFP_HIGHUSER);
487 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
488 *desc->dir_cookie, page,
489 NFS_SERVER(inode)->dtsize,
492 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
494 desc->timestamp = timestamp;
495 desc->timestamp_valid = 1;
496 if ((status = dir_decode(desc)) == 0)
497 desc->entry->prev_cookie = *desc->dir_cookie;
503 status = nfs_do_filldir(desc, dirent, filldir);
505 /* Reset read descriptor so it searches the page cache from
506 * the start upon the next call to readdir_search_pagecache() */
507 desc->page_index = 0;
508 desc->entry->cookie = desc->entry->prev_cookie = 0;
509 desc->entry->eof = 0;
511 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
515 dir_page_release(desc);
519 /* The file offset position represents the dirent entry number. A
520 last cookie cache takes care of the common case of reading the
523 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
525 struct dentry *dentry = filp->f_path.dentry;
526 struct inode *inode = dentry->d_inode;
527 nfs_readdir_descriptor_t my_desc,
529 struct nfs_entry my_entry;
531 struct nfs_fattr fattr;
534 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
535 dentry->d_parent->d_name.name, dentry->d_name.name,
536 (long long)filp->f_pos);
537 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
542 * filp->f_pos points to the dirent entry number.
543 * *desc->dir_cookie has the cookie for the next entry. We have
544 * to either find the entry with the appropriate number or
545 * revalidate the cookie.
547 memset(desc, 0, sizeof(*desc));
550 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
551 desc->decode = NFS_PROTO(inode)->decode_dirent;
552 desc->plus = NFS_USE_READDIRPLUS(inode);
554 my_entry.cookie = my_entry.prev_cookie = 0;
557 my_entry.fattr = &fattr;
558 nfs_fattr_init(&fattr);
559 desc->entry = &my_entry;
561 nfs_block_sillyrename(dentry);
562 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
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_I(inode)->flags);
582 nfs_zap_caches(inode);
584 desc->entry->eof = 0;
590 res = nfs_do_filldir(desc, dirent, filldir);
597 nfs_unblock_sillyrename(dentry);
601 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
602 dentry->d_parent->d_name.name, dentry->d_name.name,
607 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
609 struct dentry *dentry = filp->f_path.dentry;
610 struct inode *inode = dentry->d_inode;
612 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
613 dentry->d_parent->d_name.name,
617 mutex_lock(&inode->i_mutex);
620 offset += filp->f_pos;
628 if (offset != filp->f_pos) {
629 filp->f_pos = offset;
630 nfs_file_open_context(filp)->dir_cookie = 0;
633 mutex_unlock(&inode->i_mutex);
638 * All directory operations under NFS are synchronous, so fsync()
639 * is a dummy operation.
641 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
643 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
644 dentry->d_parent->d_name.name, dentry->d_name.name,
647 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
652 * nfs_force_lookup_revalidate - Mark the directory as having changed
653 * @dir - pointer to directory inode
655 * This forces the revalidation code in nfs_lookup_revalidate() to do a
656 * full lookup on all child dentries of 'dir' whenever a change occurs
657 * on the server that might have invalidated our dcache.
659 * The caller should be holding dir->i_lock
661 void nfs_force_lookup_revalidate(struct inode *dir)
663 NFS_I(dir)->cache_change_attribute = jiffies;
667 * A check for whether or not the parent directory has changed.
668 * In the case it has, we assume that the dentries are untrustworthy
669 * and may need to be looked up again.
671 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
675 if (!nfs_verify_change_attribute(dir, dentry->d_time))
677 /* Revalidate nfsi->cache_change_attribute before we declare a match */
678 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
680 if (!nfs_verify_change_attribute(dir, dentry->d_time))
686 * Return the intent data that applies to this particular path component
688 * Note that the current set of intents only apply to the very last
689 * component of the path.
690 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
692 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
694 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
696 return nd->flags & mask;
700 * Use intent information to check whether or not we're going to do
701 * an O_EXCL create using this path component.
703 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
705 if (NFS_PROTO(dir)->version == 2)
707 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
709 return (nd->intent.open.flags & O_EXCL) != 0;
713 * Inode and filehandle revalidation for lookups.
715 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
716 * or if the intent information indicates that we're about to open this
717 * particular file and the "nocto" mount flag is not set.
721 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
723 struct nfs_server *server = NFS_SERVER(inode);
725 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
728 /* VFS wants an on-the-wire revalidation */
729 if (nd->flags & LOOKUP_REVAL)
731 /* This is an open(2) */
732 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
733 !(server->flags & NFS_MOUNT_NOCTO) &&
734 (S_ISREG(inode->i_mode) ||
735 S_ISDIR(inode->i_mode)))
739 return nfs_revalidate_inode(server, inode);
741 return __nfs_revalidate_inode(server, inode);
745 * We judge how long we want to trust negative
746 * dentries by looking at the parent inode mtime.
748 * If parent mtime has changed, we revalidate, else we wait for a
749 * period corresponding to the parent's attribute cache timeout value.
752 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
753 struct nameidata *nd)
755 /* Don't revalidate a negative dentry if we're creating a new file */
756 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
758 return !nfs_check_verifier(dir, dentry);
762 * This is called every time the dcache has a lookup hit,
763 * and we should check whether we can really trust that
766 * NOTE! The hit can be a negative hit too, don't assume
769 * If the parent directory is seen to have changed, we throw out the
770 * cached dentry and do a new lookup.
772 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
776 struct dentry *parent;
778 struct nfs_fh fhandle;
779 struct nfs_fattr fattr;
781 parent = dget_parent(dentry);
783 dir = parent->d_inode;
784 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
785 inode = dentry->d_inode;
788 if (nfs_neg_need_reval(dir, dentry, nd))
793 if (is_bad_inode(inode)) {
794 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
795 __func__, dentry->d_parent->d_name.name,
796 dentry->d_name.name);
800 /* Force a full look up iff the parent directory has changed */
801 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
802 if (nfs_lookup_verify_inode(inode, nd))
807 if (NFS_STALE(inode))
810 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
813 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
815 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
818 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
822 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
823 __func__, dentry->d_parent->d_name.name,
824 dentry->d_name.name);
829 nfs_mark_for_revalidate(dir);
830 if (inode && S_ISDIR(inode->i_mode)) {
831 /* Purge readdir caches. */
832 nfs_zap_caches(inode);
833 /* If we have submounts, don't unhash ! */
834 if (have_submounts(dentry))
836 shrink_dcache_parent(dentry);
841 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
842 __func__, dentry->d_parent->d_name.name,
843 dentry->d_name.name);
848 * This is called from dput() when d_count is going to 0.
850 static int nfs_dentry_delete(struct dentry *dentry)
852 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
853 dentry->d_parent->d_name.name, dentry->d_name.name,
856 /* Unhash any dentry with a stale inode */
857 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
860 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
861 /* Unhash it, so that ->d_iput() would be called */
864 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
865 /* Unhash it, so that ancestors of killed async unlink
866 * files will be cleaned up during umount */
873 static void nfs_drop_nlink(struct inode *inode)
875 spin_lock(&inode->i_lock);
876 if (inode->i_nlink > 0)
878 spin_unlock(&inode->i_lock);
882 * Called when the dentry loses inode.
883 * We use it to clean up silly-renamed files.
885 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
887 if (S_ISDIR(inode->i_mode))
888 /* drop any readdir cache as it could easily be old */
889 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
891 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
894 nfs_complete_unlink(dentry, inode);
900 struct dentry_operations nfs_dentry_operations = {
901 .d_revalidate = nfs_lookup_revalidate,
902 .d_delete = nfs_dentry_delete,
903 .d_iput = nfs_dentry_iput,
906 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
909 struct dentry *parent;
910 struct inode *inode = NULL;
912 struct nfs_fh fhandle;
913 struct nfs_fattr fattr;
915 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
916 dentry->d_parent->d_name.name, dentry->d_name.name);
917 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
919 res = ERR_PTR(-ENAMETOOLONG);
920 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
923 res = ERR_PTR(-ENOMEM);
924 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
929 * If we're doing an exclusive create, optimize away the lookup
930 * but don't hash the dentry.
932 if (nfs_is_exclusive_create(dir, nd)) {
933 d_instantiate(dentry, NULL);
938 parent = dentry->d_parent;
939 /* Protect against concurrent sillydeletes */
940 nfs_block_sillyrename(parent);
941 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
942 if (error == -ENOENT)
945 res = ERR_PTR(error);
946 goto out_unblock_sillyrename;
948 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
949 res = (struct dentry *)inode;
951 goto out_unblock_sillyrename;
954 res = d_materialise_unique(dentry, inode);
957 goto out_unblock_sillyrename;
960 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
961 out_unblock_sillyrename:
962 nfs_unblock_sillyrename(parent);
970 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
972 struct dentry_operations nfs4_dentry_operations = {
973 .d_revalidate = nfs_open_revalidate,
974 .d_delete = nfs_dentry_delete,
975 .d_iput = nfs_dentry_iput,
979 * Use intent information to determine whether we need to substitute
980 * the NFSv4-style stateful OPEN for the LOOKUP call
982 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
984 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
986 /* NFS does not (yet) have a stateful open for directories */
987 if (nd->flags & LOOKUP_DIRECTORY)
989 /* Are we trying to write to a read only partition? */
990 if (__mnt_is_readonly(nd->path.mnt) &&
991 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
996 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
998 struct dentry *res = NULL;
1001 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1002 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1004 /* Check that we are indeed trying to open this file */
1005 if (!is_atomic_open(dir, nd))
1008 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1009 res = ERR_PTR(-ENAMETOOLONG);
1012 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1014 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1016 if (nd->intent.open.flags & O_EXCL) {
1017 d_instantiate(dentry, NULL);
1021 /* Open the file on the server */
1023 res = nfs4_atomic_open(dir, dentry, nd);
1026 error = PTR_ERR(res);
1028 /* Make a negative dentry */
1032 /* This turned out not to be a regular file */
1037 if (!(nd->intent.open.flags & O_NOFOLLOW))
1043 } else if (res != NULL)
1048 return nfs_lookup(dir, dentry, nd);
1051 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1053 struct dentry *parent = NULL;
1054 struct inode *inode = dentry->d_inode;
1056 int openflags, ret = 0;
1058 parent = dget_parent(dentry);
1059 dir = parent->d_inode;
1060 if (!is_atomic_open(dir, nd))
1062 /* We can't create new files in nfs_open_revalidate(), so we
1063 * optimize away revalidation of negative dentries.
1065 if (inode == NULL) {
1066 if (!nfs_neg_need_reval(dir, dentry, nd))
1071 /* NFS only supports OPEN on regular files */
1072 if (!S_ISREG(inode->i_mode))
1074 openflags = nd->intent.open.flags;
1075 /* We cannot do exclusive creation on a positive dentry */
1076 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1078 /* We can't create new files, or truncate existing ones here */
1079 openflags &= ~(O_CREAT|O_TRUNC);
1082 * Note: we're not holding inode->i_mutex and so may be racing with
1083 * operations that change the directory. We therefore save the
1084 * change attribute *before* we do the RPC call.
1087 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1096 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1098 return nfs_lookup_revalidate(dentry, nd);
1100 #endif /* CONFIG_NFSV4 */
1102 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1104 struct dentry *parent = desc->file->f_path.dentry;
1105 struct inode *dir = parent->d_inode;
1106 struct nfs_entry *entry = desc->entry;
1107 struct dentry *dentry, *alias;
1108 struct qstr name = {
1109 .name = entry->name,
1112 struct inode *inode;
1113 unsigned long verf = nfs_save_change_attribute(dir);
1117 if (name.name[0] == '.' && name.name[1] == '.')
1118 return dget_parent(parent);
1121 if (name.name[0] == '.')
1122 return dget(parent);
1125 spin_lock(&dir->i_lock);
1126 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1127 spin_unlock(&dir->i_lock);
1130 spin_unlock(&dir->i_lock);
1132 name.hash = full_name_hash(name.name, name.len);
1133 dentry = d_lookup(parent, &name);
1134 if (dentry != NULL) {
1135 /* Is this a positive dentry that matches the readdir info? */
1136 if (dentry->d_inode != NULL &&
1137 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1138 d_mountpoint(dentry))) {
1139 if (!desc->plus || entry->fh->size == 0)
1141 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1145 /* No, so d_drop to allow one to be created */
1149 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1151 if (name.len > NFS_SERVER(dir)->namelen)
1153 /* Note: caller is already holding the dir->i_mutex! */
1154 dentry = d_alloc(parent, &name);
1157 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1158 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1159 if (IS_ERR(inode)) {
1164 alias = d_materialise_unique(dentry, inode);
1165 if (alias != NULL) {
1173 nfs_set_verifier(dentry, verf);
1178 * Code common to create, mkdir, and mknod.
1180 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1181 struct nfs_fattr *fattr)
1183 struct dentry *parent = dget_parent(dentry);
1184 struct inode *dir = parent->d_inode;
1185 struct inode *inode;
1186 int error = -EACCES;
1190 /* We may have been initialized further down */
1191 if (dentry->d_inode)
1193 if (fhandle->size == 0) {
1194 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1198 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1199 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1200 struct nfs_server *server = NFS_SB(dentry->d_sb);
1201 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1205 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1206 error = PTR_ERR(inode);
1209 d_add(dentry, inode);
1214 nfs_mark_for_revalidate(dir);
1220 * Following a failed create operation, we drop the dentry rather
1221 * than retain a negative dentry. This avoids a problem in the event
1222 * that the operation succeeded on the server, but an error in the
1223 * reply path made it appear to have failed.
1225 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1226 struct nameidata *nd)
1232 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1233 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1235 attr.ia_mode = mode;
1236 attr.ia_valid = ATTR_MODE;
1238 if ((nd->flags & LOOKUP_CREATE) != 0)
1239 open_flags = nd->intent.open.flags;
1242 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1254 * See comments for nfs_proc_create regarding failed operations.
1257 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1262 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1263 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1265 if (!new_valid_dev(rdev))
1268 attr.ia_mode = mode;
1269 attr.ia_valid = ATTR_MODE;
1272 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1284 * See comments for nfs_proc_create regarding failed operations.
1286 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1291 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1292 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1294 attr.ia_valid = ATTR_MODE;
1295 attr.ia_mode = mode | S_IFDIR;
1298 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1309 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1311 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1315 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1319 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1320 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1323 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1324 /* Ensure the VFS deletes this inode */
1325 if (error == 0 && dentry->d_inode != NULL)
1326 clear_nlink(dentry->d_inode);
1327 else if (error == -ENOENT)
1328 nfs_dentry_handle_enoent(dentry);
1334 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1336 static unsigned int sillycounter;
1337 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1338 const int countersize = sizeof(sillycounter)*2;
1339 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1342 struct dentry *sdentry;
1345 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1346 dentry->d_parent->d_name.name, dentry->d_name.name,
1347 atomic_read(&dentry->d_count));
1348 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1351 * We don't allow a dentry to be silly-renamed twice.
1354 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1357 sprintf(silly, ".nfs%*.*Lx",
1358 fileidsize, fileidsize,
1359 (unsigned long long)NFS_FILEID(dentry->d_inode));
1361 /* Return delegation in anticipation of the rename */
1362 nfs_inode_return_delegation(dentry->d_inode);
1366 char *suffix = silly + slen - countersize;
1370 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1372 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1373 dentry->d_name.name, silly);
1375 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1377 * N.B. Better to return EBUSY here ... it could be
1378 * dangerous to delete the file while it's in use.
1380 if (IS_ERR(sdentry))
1382 } while(sdentry->d_inode != NULL); /* need negative lookup */
1384 qsilly.name = silly;
1385 qsilly.len = strlen(silly);
1386 if (dentry->d_inode) {
1387 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1389 nfs_mark_for_revalidate(dentry->d_inode);
1391 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1394 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1395 d_move(dentry, sdentry);
1396 error = nfs_async_unlink(dir, dentry);
1397 /* If we return 0 we don't unlink */
1405 * Remove a file after making sure there are no pending writes,
1406 * and after checking that the file has only one user.
1408 * We invalidate the attribute cache and free the inode prior to the operation
1409 * to avoid possible races if the server reuses the inode.
1411 static int nfs_safe_remove(struct dentry *dentry)
1413 struct inode *dir = dentry->d_parent->d_inode;
1414 struct inode *inode = dentry->d_inode;
1417 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1418 dentry->d_parent->d_name.name, dentry->d_name.name);
1420 /* If the dentry was sillyrenamed, we simply call d_delete() */
1421 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1426 if (inode != NULL) {
1427 nfs_inode_return_delegation(inode);
1428 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1429 /* The VFS may want to delete this inode */
1431 nfs_drop_nlink(inode);
1432 nfs_mark_for_revalidate(inode);
1434 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1435 if (error == -ENOENT)
1436 nfs_dentry_handle_enoent(dentry);
1441 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1442 * belongs to an active ".nfs..." file and we return -EBUSY.
1444 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1446 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1449 int need_rehash = 0;
1451 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1452 dir->i_ino, dentry->d_name.name);
1455 spin_lock(&dcache_lock);
1456 spin_lock(&dentry->d_lock);
1457 if (atomic_read(&dentry->d_count) > 1) {
1458 spin_unlock(&dentry->d_lock);
1459 spin_unlock(&dcache_lock);
1460 /* Start asynchronous writeout of the inode */
1461 write_inode_now(dentry->d_inode, 0);
1462 error = nfs_sillyrename(dir, dentry);
1466 if (!d_unhashed(dentry)) {
1470 spin_unlock(&dentry->d_lock);
1471 spin_unlock(&dcache_lock);
1472 error = nfs_safe_remove(dentry);
1473 if (!error || error == -ENOENT) {
1474 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1475 } else if (need_rehash)
1482 * To create a symbolic link, most file systems instantiate a new inode,
1483 * add a page to it containing the path, then write it out to the disk
1484 * using prepare_write/commit_write.
1486 * Unfortunately the NFS client can't create the in-core inode first
1487 * because it needs a file handle to create an in-core inode (see
1488 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1489 * symlink request has completed on the server.
1491 * So instead we allocate a raw page, copy the symname into it, then do
1492 * the SYMLINK request with the page as the buffer. If it succeeds, we
1493 * now have a new file handle and can instantiate an in-core NFS inode
1494 * and move the raw page into its mapping.
1496 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1498 struct pagevec lru_pvec;
1502 unsigned int pathlen = strlen(symname);
1505 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1506 dir->i_ino, dentry->d_name.name, symname);
1508 if (pathlen > PAGE_SIZE)
1509 return -ENAMETOOLONG;
1511 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1512 attr.ia_valid = ATTR_MODE;
1516 page = alloc_page(GFP_HIGHUSER);
1522 kaddr = kmap_atomic(page, KM_USER0);
1523 memcpy(kaddr, symname, pathlen);
1524 if (pathlen < PAGE_SIZE)
1525 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1526 kunmap_atomic(kaddr, KM_USER0);
1528 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1530 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1531 dir->i_sb->s_id, dir->i_ino,
1532 dentry->d_name.name, symname, error);
1540 * No big deal if we can't add this page to the page cache here.
1541 * READLINK will get the missing page from the server if needed.
1543 pagevec_init(&lru_pvec, 0);
1544 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1546 pagevec_add(&lru_pvec, page);
1547 pagevec_lru_add(&lru_pvec);
1548 SetPageUptodate(page);
1558 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1560 struct inode *inode = old_dentry->d_inode;
1563 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1564 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1565 dentry->d_parent->d_name.name, dentry->d_name.name);
1569 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1571 atomic_inc(&inode->i_count);
1572 d_add(dentry, inode);
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 nfs_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 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1677 new_dir, &new_dentry->d_name);
1678 nfs_mark_for_revalidate(old_inode);
1683 d_move(old_dentry, new_dentry);
1684 nfs_set_verifier(new_dentry,
1685 nfs_save_change_attribute(new_dir));
1686 } else if (error == -ENOENT)
1687 nfs_dentry_handle_enoent(old_dentry);
1689 /* new dentry created? */
1696 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1697 static LIST_HEAD(nfs_access_lru_list);
1698 static atomic_long_t nfs_access_nr_entries;
1700 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1702 put_rpccred(entry->cred);
1704 smp_mb__before_atomic_dec();
1705 atomic_long_dec(&nfs_access_nr_entries);
1706 smp_mb__after_atomic_dec();
1709 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1712 struct nfs_inode *nfsi;
1713 struct nfs_access_entry *cache;
1716 spin_lock(&nfs_access_lru_lock);
1717 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1718 struct rw_semaphore *s_umount;
1719 struct inode *inode;
1721 if (nr_to_scan-- == 0)
1723 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1724 if (!down_read_trylock(s_umount))
1726 inode = igrab(&nfsi->vfs_inode);
1727 if (inode == NULL) {
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);
1752 spin_unlock(&nfs_access_lru_lock);
1753 while (!list_empty(&head)) {
1754 cache = list_entry(head.next, struct nfs_access_entry, lru);
1755 list_del(&cache->lru);
1756 nfs_access_free_entry(cache);
1758 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1761 static void __nfs_access_zap_cache(struct inode *inode)
1763 struct nfs_inode *nfsi = NFS_I(inode);
1764 struct rb_root *root_node = &nfsi->access_cache;
1765 struct rb_node *n, *dispose = NULL;
1766 struct nfs_access_entry *entry;
1768 /* Unhook entries from the cache */
1769 while ((n = rb_first(root_node)) != NULL) {
1770 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1771 rb_erase(n, root_node);
1772 list_del(&entry->lru);
1773 n->rb_left = dispose;
1776 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1777 spin_unlock(&inode->i_lock);
1779 /* Now kill them all! */
1780 while (dispose != NULL) {
1782 dispose = n->rb_left;
1783 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1787 void nfs_access_zap_cache(struct inode *inode)
1789 /* Remove from global LRU init */
1790 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1791 spin_lock(&nfs_access_lru_lock);
1792 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1793 spin_unlock(&nfs_access_lru_lock);
1796 spin_lock(&inode->i_lock);
1797 /* This will release the spinlock */
1798 __nfs_access_zap_cache(inode);
1801 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1803 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1804 struct nfs_access_entry *entry;
1807 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1809 if (cred < entry->cred)
1811 else if (cred > entry->cred)
1819 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1821 struct nfs_inode *nfsi = NFS_I(inode);
1822 struct nfs_access_entry *cache;
1825 spin_lock(&inode->i_lock);
1826 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1828 cache = nfs_access_search_rbtree(inode, cred);
1831 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1833 res->jiffies = cache->jiffies;
1834 res->cred = cache->cred;
1835 res->mask = cache->mask;
1836 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1839 spin_unlock(&inode->i_lock);
1842 rb_erase(&cache->rb_node, &nfsi->access_cache);
1843 list_del(&cache->lru);
1844 spin_unlock(&inode->i_lock);
1845 nfs_access_free_entry(cache);
1848 /* This will release the spinlock */
1849 __nfs_access_zap_cache(inode);
1853 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1855 struct nfs_inode *nfsi = NFS_I(inode);
1856 struct rb_root *root_node = &nfsi->access_cache;
1857 struct rb_node **p = &root_node->rb_node;
1858 struct rb_node *parent = NULL;
1859 struct nfs_access_entry *entry;
1861 spin_lock(&inode->i_lock);
1862 while (*p != NULL) {
1864 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1866 if (set->cred < entry->cred)
1867 p = &parent->rb_left;
1868 else if (set->cred > entry->cred)
1869 p = &parent->rb_right;
1873 rb_link_node(&set->rb_node, parent, p);
1874 rb_insert_color(&set->rb_node, root_node);
1875 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1876 spin_unlock(&inode->i_lock);
1879 rb_replace_node(parent, &set->rb_node, root_node);
1880 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1881 list_del(&entry->lru);
1882 spin_unlock(&inode->i_lock);
1883 nfs_access_free_entry(entry);
1886 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1888 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1891 RB_CLEAR_NODE(&cache->rb_node);
1892 cache->jiffies = set->jiffies;
1893 cache->cred = get_rpccred(set->cred);
1894 cache->mask = set->mask;
1896 nfs_access_add_rbtree(inode, cache);
1898 /* Update accounting */
1899 smp_mb__before_atomic_inc();
1900 atomic_long_inc(&nfs_access_nr_entries);
1901 smp_mb__after_atomic_inc();
1903 /* Add inode to global LRU list */
1904 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1905 spin_lock(&nfs_access_lru_lock);
1906 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1907 spin_unlock(&nfs_access_lru_lock);
1911 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1913 struct nfs_access_entry cache;
1916 status = nfs_access_get_cached(inode, cred, &cache);
1920 /* Be clever: ask server to check for all possible rights */
1921 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1923 cache.jiffies = jiffies;
1924 status = NFS_PROTO(inode)->access(inode, &cache);
1927 nfs_access_add_cache(inode, &cache);
1929 if ((cache.mask & mask) == mask)
1934 static int nfs_open_permission_mask(int openflags)
1938 if (openflags & FMODE_READ)
1940 if (openflags & FMODE_WRITE)
1942 if (openflags & FMODE_EXEC)
1947 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1949 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1952 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1954 struct rpc_cred *cred;
1957 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1961 /* Is this sys_access() ? */
1962 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1965 switch (inode->i_mode & S_IFMT) {
1969 /* NFSv4 has atomic_open... */
1970 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1972 && (nd->flags & LOOKUP_OPEN))
1977 * Optimize away all write operations, since the server
1978 * will check permissions when we perform the op.
1980 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1987 if (!NFS_PROTO(inode)->access)
1990 cred = rpc_lookup_cred();
1991 if (!IS_ERR(cred)) {
1992 res = nfs_do_access(inode, cred, mask);
1995 res = PTR_ERR(cred);
1998 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1999 inode->i_sb->s_id, inode->i_ino, mask, res);
2002 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2004 res = generic_permission(inode, mask, NULL);
2011 * version-control: t
2012 * kept-new-versions: 5