4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
39 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode *, struct file *);
46 static int nfs_readdir(struct file *, void *, filldir_t);
47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49 static int nfs_mkdir(struct inode *, struct dentry *, int);
50 static int nfs_rmdir(struct inode *, struct dentry *);
51 static int nfs_unlink(struct inode *, struct dentry *);
52 static int nfs_symlink(struct inode *, struct dentry *, const char *);
53 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55 static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57 static int nfs_fsync_dir(struct file *, struct dentry *, int);
58 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 const struct file_operations nfs_dir_operations = {
61 .llseek = nfs_llseek_dir,
62 .read = generic_read_dir,
63 .readdir = nfs_readdir,
65 .release = nfs_release,
66 .fsync = nfs_fsync_dir,
69 const struct inode_operations nfs_dir_inode_operations = {
74 .symlink = nfs_symlink,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
85 const struct inode_operations nfs3_dir_inode_operations = {
90 .symlink = nfs_symlink,
95 .permission = nfs_permission,
96 .getattr = nfs_getattr,
97 .setattr = nfs_setattr,
98 .listxattr = nfs3_listxattr,
99 .getxattr = nfs3_getxattr,
100 .setxattr = nfs3_setxattr,
101 .removexattr = nfs3_removexattr,
103 #endif /* CONFIG_NFS_V3 */
107 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
108 const struct inode_operations nfs4_dir_inode_operations = {
109 .create = nfs_create,
110 .lookup = nfs_atomic_lookup,
112 .unlink = nfs_unlink,
113 .symlink = nfs_symlink,
117 .rename = nfs_rename,
118 .permission = nfs_permission,
119 .getattr = nfs_getattr,
120 .setattr = nfs_setattr,
121 .getxattr = nfs4_getxattr,
122 .setxattr = nfs4_setxattr,
123 .listxattr = nfs4_listxattr,
126 #endif /* CONFIG_NFS_V4 */
132 nfs_opendir(struct inode *inode, struct file *filp)
136 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
137 filp->f_path.dentry->d_parent->d_name.name,
138 filp->f_path.dentry->d_name.name);
140 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
142 /* Call generic open code in order to cache credentials */
143 res = nfs_open(inode, filp);
147 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
151 unsigned long page_index;
154 loff_t current_index;
155 struct nfs_entry *entry;
156 decode_dirent_t decode;
158 unsigned long timestamp;
159 unsigned long gencount;
161 } nfs_readdir_descriptor_t;
163 /* Now we cache directories properly, by stuffing the dirent
164 * data directly in the page cache.
166 * Inode invalidation due to refresh etc. takes care of
167 * _everything_, no sloppy entry flushing logic, no extraneous
168 * copying, network direct to page cache, the way it was meant
171 * NOTE: Dirent information verification is done always by the
172 * page-in of the RPC reply, nowhere else, this simplies
173 * things substantially.
176 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
178 struct file *file = desc->file;
179 struct inode *inode = file->f_path.dentry->d_inode;
180 struct rpc_cred *cred = nfs_file_cred(file);
181 unsigned long timestamp, gencount;
184 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
185 __func__, (long long)desc->entry->cookie,
190 gencount = nfs_inc_attr_generation_counter();
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->gencount = gencount;
205 desc->timestamp_valid = 1;
206 SetPageUptodate(page);
207 /* Ensure consistent page alignment of the data.
208 * Note: assumes we have exclusive access to this mapping either
209 * through inode->i_mutex or some other mechanism.
211 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
212 /* Should never happen */
213 nfs_zap_mapping(inode, inode->i_mapping);
223 int dir_decode(nfs_readdir_descriptor_t *desc)
225 __be32 *p = desc->ptr;
226 p = desc->decode(p, desc->entry, desc->plus);
230 if (desc->timestamp_valid) {
231 desc->entry->fattr->time_start = desc->timestamp;
232 desc->entry->fattr->gencount = desc->gencount;
234 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
239 void dir_page_release(nfs_readdir_descriptor_t *desc)
242 page_cache_release(desc->page);
248 * Given a pointer to a buffer that has already been filled by a call
249 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
251 * If the end of the buffer has been reached, return -EAGAIN, if not,
252 * return the offset within the buffer of the next entry to be
256 int find_dirent(nfs_readdir_descriptor_t *desc)
258 struct nfs_entry *entry = desc->entry;
262 while((status = dir_decode(desc)) == 0) {
263 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
264 __func__, (unsigned long long)entry->cookie);
265 if (entry->prev_cookie == *desc->dir_cookie)
267 if (loop_count++ > 200) {
276 * Given a pointer to a buffer that has already been filled by a call
277 * to readdir, find the entry at offset 'desc->file->f_pos'.
279 * If the end of the buffer has been reached, return -EAGAIN, if not,
280 * return the offset within the buffer of the next entry to be
284 int find_dirent_index(nfs_readdir_descriptor_t *desc)
286 struct nfs_entry *entry = desc->entry;
291 status = dir_decode(desc);
295 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
296 (unsigned long long)entry->cookie, desc->current_index);
298 if (desc->file->f_pos == desc->current_index) {
299 *desc->dir_cookie = entry->cookie;
302 desc->current_index++;
303 if (loop_count++ > 200) {
312 * Find the given page, and call find_dirent() or find_dirent_index in
313 * order to try to return the next entry.
316 int find_dirent_page(nfs_readdir_descriptor_t *desc)
318 struct inode *inode = desc->file->f_path.dentry->d_inode;
322 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
323 __func__, desc->page_index,
324 (long long) *desc->dir_cookie);
326 /* If we find the page in the page_cache, we cannot be sure
327 * how fresh the data is, so we will ignore readdir_plus attributes.
329 desc->timestamp_valid = 0;
330 page = read_cache_page(inode->i_mapping, desc->page_index,
331 (filler_t *)nfs_readdir_filler, desc);
333 status = PTR_ERR(page);
337 /* NOTE: Someone else may have changed the READDIRPLUS flag */
339 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
340 if (*desc->dir_cookie != 0)
341 status = find_dirent(desc);
343 status = find_dirent_index(desc);
345 dir_page_release(desc);
347 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
352 * Recurse through the page cache pages, and return a
353 * filled nfs_entry structure of the next directory entry if possible.
355 * The target for the search is '*desc->dir_cookie' if non-0,
356 * 'desc->file->f_pos' otherwise
359 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
364 /* Always search-by-index from the beginning of the cache */
365 if (*desc->dir_cookie == 0) {
366 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
367 (long long)desc->file->f_pos);
368 desc->page_index = 0;
369 desc->entry->cookie = desc->entry->prev_cookie = 0;
370 desc->entry->eof = 0;
371 desc->current_index = 0;
373 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
374 (unsigned long long)*desc->dir_cookie);
377 res = find_dirent_page(desc);
380 /* Align to beginning of next page */
382 if (loop_count++ > 200) {
388 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
392 static inline unsigned int dt_type(struct inode *inode)
394 return (inode->i_mode >> 12) & 15;
397 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
400 * Once we've found the start of the dirent within a page: fill 'er up...
403 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
406 struct file *file = desc->file;
407 struct nfs_entry *entry = desc->entry;
408 struct dentry *dentry = NULL;
413 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
414 (unsigned long long)entry->cookie);
417 unsigned d_type = DT_UNKNOWN;
418 /* Note: entry->prev_cookie contains the cookie for
419 * retrieving the current dirent on the server */
422 /* Get a dentry if we have one */
425 dentry = nfs_readdir_lookup(desc);
427 /* Use readdirplus info */
428 if (dentry != NULL && dentry->d_inode != NULL) {
429 d_type = dt_type(dentry->d_inode);
430 fileid = NFS_FILEID(dentry->d_inode);
433 res = filldir(dirent, entry->name, entry->len,
434 file->f_pos, nfs_compat_user_ino64(fileid),
439 *desc->dir_cookie = entry->cookie;
440 if (dir_decode(desc) != 0) {
444 if (loop_count++ > 200) {
449 dir_page_release(desc);
452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
453 (unsigned long long)*desc->dir_cookie, res);
458 * If we cannot find a cookie in our cache, we suspect that this is
459 * because it points to a deleted file, so we ask the server to return
460 * whatever it thinks is the next entry. We then feed this to filldir.
461 * If all goes well, we should then be able to find our way round the
462 * cache on the next call to readdir_search_pagecache();
464 * NOTE: we cannot add the anonymous page to the pagecache because
465 * the data it contains might not be page aligned. Besides,
466 * we should already have a complete representation of the
467 * directory in the page cache by the time we get here.
470 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
473 struct file *file = desc->file;
474 struct inode *inode = file->f_path.dentry->d_inode;
475 struct rpc_cred *cred = nfs_file_cred(file);
476 struct page *page = NULL;
478 unsigned long timestamp, gencount;
480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
481 (unsigned long long)*desc->dir_cookie);
483 page = alloc_page(GFP_HIGHUSER);
489 gencount = nfs_inc_attr_generation_counter();
490 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
491 *desc->dir_cookie, page,
492 NFS_SERVER(inode)->dtsize,
495 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
497 desc->timestamp = timestamp;
498 desc->gencount = gencount;
499 desc->timestamp_valid = 1;
500 if ((status = dir_decode(desc)) == 0)
501 desc->entry->prev_cookie = *desc->dir_cookie;
507 status = nfs_do_filldir(desc, dirent, filldir);
509 /* Reset read descriptor so it searches the page cache from
510 * the start upon the next call to readdir_search_pagecache() */
511 desc->page_index = 0;
512 desc->entry->cookie = desc->entry->prev_cookie = 0;
513 desc->entry->eof = 0;
515 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
519 dir_page_release(desc);
523 /* The file offset position represents the dirent entry number. A
524 last cookie cache takes care of the common case of reading the
527 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
529 struct dentry *dentry = filp->f_path.dentry;
530 struct inode *inode = dentry->d_inode;
531 nfs_readdir_descriptor_t my_desc,
533 struct nfs_entry my_entry;
535 struct nfs_fattr fattr;
538 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
539 dentry->d_parent->d_name.name, dentry->d_name.name,
540 (long long)filp->f_pos);
541 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
544 * filp->f_pos points to the dirent entry number.
545 * *desc->dir_cookie has the cookie for the next entry. We have
546 * to either find the entry with the appropriate number or
547 * revalidate the cookie.
549 memset(desc, 0, sizeof(*desc));
552 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
553 desc->decode = NFS_PROTO(inode)->decode_dirent;
554 desc->plus = NFS_USE_READDIRPLUS(inode);
556 my_entry.cookie = my_entry.prev_cookie = 0;
559 my_entry.fattr = &fattr;
560 nfs_fattr_init(&fattr);
561 desc->entry = &my_entry;
563 nfs_block_sillyrename(dentry);
564 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
568 while(!desc->entry->eof) {
569 res = readdir_search_pagecache(desc);
571 if (res == -EBADCOOKIE) {
572 /* This means either end of directory */
573 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
574 /* Or that the server has 'lost' a cookie */
575 res = uncached_readdir(desc, dirent, filldir);
582 if (res == -ETOOSMALL && desc->plus) {
583 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
584 nfs_zap_caches(inode);
586 desc->entry->eof = 0;
592 res = nfs_do_filldir(desc, dirent, filldir);
599 nfs_unblock_sillyrename(dentry);
602 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
603 dentry->d_parent->d_name.name, dentry->d_name.name,
608 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
610 struct dentry *dentry = filp->f_path.dentry;
611 struct inode *inode = dentry->d_inode;
613 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
614 dentry->d_parent->d_name.name,
618 mutex_lock(&inode->i_mutex);
621 offset += filp->f_pos;
629 if (offset != filp->f_pos) {
630 filp->f_pos = offset;
631 nfs_file_open_context(filp)->dir_cookie = 0;
634 mutex_unlock(&inode->i_mutex);
639 * All directory operations under NFS are synchronous, so fsync()
640 * is a dummy operation.
642 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
644 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
645 dentry->d_parent->d_name.name, dentry->d_name.name,
648 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
653 * nfs_force_lookup_revalidate - Mark the directory as having changed
654 * @dir - pointer to directory inode
656 * This forces the revalidation code in nfs_lookup_revalidate() to do a
657 * full lookup on all child dentries of 'dir' whenever a change occurs
658 * on the server that might have invalidated our dcache.
660 * The caller should be holding dir->i_lock
662 void nfs_force_lookup_revalidate(struct inode *dir)
664 NFS_I(dir)->cache_change_attribute++;
668 * A check for whether or not the parent directory has changed.
669 * In the case it has, we assume that the dentries are untrustworthy
670 * and may need to be looked up again.
672 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
676 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
678 if (!nfs_verify_change_attribute(dir, dentry->d_time))
680 /* Revalidate nfsi->cache_change_attribute before we declare a match */
681 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
683 if (!nfs_verify_change_attribute(dir, dentry->d_time))
689 * Return the intent data that applies to this particular path component
691 * Note that the current set of intents only apply to the very last
692 * component of the path.
693 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
695 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
697 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
699 return nd->flags & mask;
703 * Use intent information to check whether or not we're going to do
704 * an O_EXCL create using this path component.
706 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
708 if (NFS_PROTO(dir)->version == 2)
710 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
712 return (nd->intent.open.flags & O_EXCL) != 0;
716 * Inode and filehandle revalidation for lookups.
718 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
719 * or if the intent information indicates that we're about to open this
720 * particular file and the "nocto" mount flag is not set.
724 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
726 struct nfs_server *server = NFS_SERVER(inode);
728 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
731 /* VFS wants an on-the-wire revalidation */
732 if (nd->flags & LOOKUP_REVAL)
734 /* This is an open(2) */
735 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
736 !(server->flags & NFS_MOUNT_NOCTO) &&
737 (S_ISREG(inode->i_mode) ||
738 S_ISDIR(inode->i_mode)))
742 return nfs_revalidate_inode(server, inode);
744 return __nfs_revalidate_inode(server, inode);
748 * We judge how long we want to trust negative
749 * dentries by looking at the parent inode mtime.
751 * If parent mtime has changed, we revalidate, else we wait for a
752 * period corresponding to the parent's attribute cache timeout value.
755 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
756 struct nameidata *nd)
758 /* Don't revalidate a negative dentry if we're creating a new file */
759 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
761 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
763 return !nfs_check_verifier(dir, dentry);
767 * This is called every time the dcache has a lookup hit,
768 * and we should check whether we can really trust that
771 * NOTE! The hit can be a negative hit too, don't assume
774 * If the parent directory is seen to have changed, we throw out the
775 * cached dentry and do a new lookup.
777 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
781 struct dentry *parent;
783 struct nfs_fh fhandle;
784 struct nfs_fattr fattr;
786 parent = dget_parent(dentry);
787 dir = parent->d_inode;
788 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
789 inode = dentry->d_inode;
792 if (nfs_neg_need_reval(dir, dentry, nd))
797 if (is_bad_inode(inode)) {
798 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
799 __func__, dentry->d_parent->d_name.name,
800 dentry->d_name.name);
804 /* Force a full look up iff the parent directory has changed */
805 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
806 if (nfs_lookup_verify_inode(inode, nd))
811 if (NFS_STALE(inode))
814 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
817 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
819 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
822 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
825 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
826 __func__, dentry->d_parent->d_name.name,
827 dentry->d_name.name);
832 nfs_mark_for_revalidate(dir);
833 if (inode && S_ISDIR(inode->i_mode)) {
834 /* Purge readdir caches. */
835 nfs_zap_caches(inode);
836 /* If we have submounts, don't unhash ! */
837 if (have_submounts(dentry))
839 shrink_dcache_parent(dentry);
843 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
844 __func__, dentry->d_parent->d_name.name,
845 dentry->d_name.name);
850 * This is called from dput() when d_count is going to 0.
852 static int nfs_dentry_delete(struct dentry *dentry)
854 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
855 dentry->d_parent->d_name.name, dentry->d_name.name,
858 /* Unhash any dentry with a stale inode */
859 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
862 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
863 /* Unhash it, so that ->d_iput() would be called */
866 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
867 /* Unhash it, so that ancestors of killed async unlink
868 * files will be cleaned up during umount */
875 static void nfs_drop_nlink(struct inode *inode)
877 spin_lock(&inode->i_lock);
878 if (inode->i_nlink > 0)
880 spin_unlock(&inode->i_lock);
884 * Called when the dentry loses inode.
885 * We use it to clean up silly-renamed files.
887 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
889 if (S_ISDIR(inode->i_mode))
890 /* drop any readdir cache as it could easily be old */
891 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
893 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
895 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;
927 * If we're doing an exclusive create, optimize away the lookup
928 * but don't hash the dentry.
930 if (nfs_is_exclusive_create(dir, nd)) {
931 d_instantiate(dentry, NULL);
936 parent = dentry->d_parent;
937 /* Protect against concurrent sillydeletes */
938 nfs_block_sillyrename(parent);
939 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
940 if (error == -ENOENT)
943 res = ERR_PTR(error);
944 goto out_unblock_sillyrename;
946 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
947 res = (struct dentry *)inode;
949 goto out_unblock_sillyrename;
952 res = d_materialise_unique(dentry, inode);
955 goto out_unblock_sillyrename;
958 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
959 out_unblock_sillyrename:
960 nfs_unblock_sillyrename(parent);
966 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
968 struct dentry_operations nfs4_dentry_operations = {
969 .d_revalidate = nfs_open_revalidate,
970 .d_delete = nfs_dentry_delete,
971 .d_iput = nfs_dentry_iput,
975 * Use intent information to determine whether we need to substitute
976 * the NFSv4-style stateful OPEN for the LOOKUP call
978 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
980 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
982 /* NFS does not (yet) have a stateful open for directories */
983 if (nd->flags & LOOKUP_DIRECTORY)
985 /* Are we trying to write to a read only partition? */
986 if (__mnt_is_readonly(nd->path.mnt) &&
987 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
992 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
994 struct dentry *res = NULL;
997 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
998 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1000 /* Check that we are indeed trying to open this file */
1001 if (!is_atomic_open(dir, nd))
1004 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1005 res = ERR_PTR(-ENAMETOOLONG);
1008 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1010 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1012 if (nd->intent.open.flags & O_EXCL) {
1013 d_instantiate(dentry, NULL);
1017 /* Open the file on the server */
1018 res = nfs4_atomic_open(dir, dentry, nd);
1020 error = PTR_ERR(res);
1022 /* Make a negative dentry */
1026 /* This turned out not to be a regular file */
1031 if (!(nd->intent.open.flags & O_NOFOLLOW))
1037 } else if (res != NULL)
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 int openflags, ret = 0;
1052 parent = dget_parent(dentry);
1053 dir = parent->d_inode;
1054 if (!is_atomic_open(dir, nd))
1056 /* We can't create new files in nfs_open_revalidate(), so we
1057 * optimize away revalidation of negative dentries.
1059 if (inode == NULL) {
1060 if (!nfs_neg_need_reval(dir, dentry, nd))
1065 /* NFS only supports OPEN on regular files */
1066 if (!S_ISREG(inode->i_mode))
1068 openflags = nd->intent.open.flags;
1069 /* We cannot do exclusive creation on a positive dentry */
1070 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1072 /* We can't create new files, or truncate existing ones here */
1073 openflags &= ~(O_CREAT|O_TRUNC);
1076 * Note: we're not holding inode->i_mutex and so may be racing with
1077 * operations that change the directory. We therefore save the
1078 * change attribute *before* we do the RPC call.
1080 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1088 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1090 return nfs_lookup_revalidate(dentry, nd);
1092 #endif /* CONFIG_NFSV4 */
1094 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1096 struct dentry *parent = desc->file->f_path.dentry;
1097 struct inode *dir = parent->d_inode;
1098 struct nfs_entry *entry = desc->entry;
1099 struct dentry *dentry, *alias;
1100 struct qstr name = {
1101 .name = entry->name,
1104 struct inode *inode;
1105 unsigned long verf = nfs_save_change_attribute(dir);
1109 if (name.name[0] == '.' && name.name[1] == '.')
1110 return dget_parent(parent);
1113 if (name.name[0] == '.')
1114 return dget(parent);
1117 spin_lock(&dir->i_lock);
1118 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1119 spin_unlock(&dir->i_lock);
1122 spin_unlock(&dir->i_lock);
1124 name.hash = full_name_hash(name.name, name.len);
1125 dentry = d_lookup(parent, &name);
1126 if (dentry != NULL) {
1127 /* Is this a positive dentry that matches the readdir info? */
1128 if (dentry->d_inode != NULL &&
1129 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1130 d_mountpoint(dentry))) {
1131 if (!desc->plus || entry->fh->size == 0)
1133 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1137 /* No, so d_drop to allow one to be created */
1141 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1143 if (name.len > NFS_SERVER(dir)->namelen)
1145 /* Note: caller is already holding the dir->i_mutex! */
1146 dentry = d_alloc(parent, &name);
1149 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1150 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1151 if (IS_ERR(inode)) {
1156 alias = d_materialise_unique(dentry, inode);
1157 if (alias != NULL) {
1165 nfs_set_verifier(dentry, verf);
1170 * Code common to create, mkdir, and mknod.
1172 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1173 struct nfs_fattr *fattr)
1175 struct dentry *parent = dget_parent(dentry);
1176 struct inode *dir = parent->d_inode;
1177 struct inode *inode;
1178 int error = -EACCES;
1182 /* We may have been initialized further down */
1183 if (dentry->d_inode)
1185 if (fhandle->size == 0) {
1186 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1190 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1191 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1192 struct nfs_server *server = NFS_SB(dentry->d_sb);
1193 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1197 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1198 error = PTR_ERR(inode);
1201 d_add(dentry, inode);
1206 nfs_mark_for_revalidate(dir);
1212 * Following a failed create operation, we drop the dentry rather
1213 * than retain a negative dentry. This avoids a problem in the event
1214 * that the operation succeeded on the server, but an error in the
1215 * reply path made it appear to have failed.
1217 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1218 struct nameidata *nd)
1224 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1225 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1227 attr.ia_mode = mode;
1228 attr.ia_valid = ATTR_MODE;
1230 if ((nd->flags & LOOKUP_CREATE) != 0)
1231 open_flags = nd->intent.open.flags;
1233 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
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;
1260 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1270 * See comments for nfs_proc_create regarding failed operations.
1272 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1277 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1278 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1280 attr.ia_valid = ATTR_MODE;
1281 attr.ia_mode = mode | S_IFDIR;
1283 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1292 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1294 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1298 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1302 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1303 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1305 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1306 /* Ensure the VFS deletes this inode */
1307 if (error == 0 && dentry->d_inode != NULL)
1308 clear_nlink(dentry->d_inode);
1309 else if (error == -ENOENT)
1310 nfs_dentry_handle_enoent(dentry);
1315 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1317 static unsigned int sillycounter;
1318 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1319 const int countersize = sizeof(sillycounter)*2;
1320 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1323 struct dentry *sdentry;
1326 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1327 dentry->d_parent->d_name.name, dentry->d_name.name,
1328 atomic_read(&dentry->d_count));
1329 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1332 * We don't allow a dentry to be silly-renamed twice.
1335 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1338 sprintf(silly, ".nfs%*.*Lx",
1339 fileidsize, fileidsize,
1340 (unsigned long long)NFS_FILEID(dentry->d_inode));
1342 /* Return delegation in anticipation of the rename */
1343 nfs_inode_return_delegation(dentry->d_inode);
1347 char *suffix = silly + slen - countersize;
1351 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1353 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1354 dentry->d_name.name, silly);
1356 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1358 * N.B. Better to return EBUSY here ... it could be
1359 * dangerous to delete the file while it's in use.
1361 if (IS_ERR(sdentry))
1363 } while(sdentry->d_inode != NULL); /* need negative lookup */
1365 qsilly.name = silly;
1366 qsilly.len = strlen(silly);
1367 if (dentry->d_inode) {
1368 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1370 nfs_mark_for_revalidate(dentry->d_inode);
1372 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1375 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1376 d_move(dentry, sdentry);
1377 error = nfs_async_unlink(dir, dentry);
1378 /* If we return 0 we don't unlink */
1386 * Remove a file after making sure there are no pending writes,
1387 * and after checking that the file has only one user.
1389 * We invalidate the attribute cache and free the inode prior to the operation
1390 * to avoid possible races if the server reuses the inode.
1392 static int nfs_safe_remove(struct dentry *dentry)
1394 struct inode *dir = dentry->d_parent->d_inode;
1395 struct inode *inode = dentry->d_inode;
1398 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1399 dentry->d_parent->d_name.name, dentry->d_name.name);
1401 /* If the dentry was sillyrenamed, we simply call d_delete() */
1402 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1407 if (inode != NULL) {
1408 nfs_inode_return_delegation(inode);
1409 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1410 /* The VFS may want to delete this inode */
1412 nfs_drop_nlink(inode);
1413 nfs_mark_for_revalidate(inode);
1415 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1416 if (error == -ENOENT)
1417 nfs_dentry_handle_enoent(dentry);
1422 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1423 * belongs to an active ".nfs..." file and we return -EBUSY.
1425 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1427 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1430 int need_rehash = 0;
1432 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1433 dir->i_ino, dentry->d_name.name);
1435 spin_lock(&dcache_lock);
1436 spin_lock(&dentry->d_lock);
1437 if (atomic_read(&dentry->d_count) > 1) {
1438 spin_unlock(&dentry->d_lock);
1439 spin_unlock(&dcache_lock);
1440 /* Start asynchronous writeout of the inode */
1441 write_inode_now(dentry->d_inode, 0);
1442 error = nfs_sillyrename(dir, dentry);
1445 if (!d_unhashed(dentry)) {
1449 spin_unlock(&dentry->d_lock);
1450 spin_unlock(&dcache_lock);
1451 error = nfs_safe_remove(dentry);
1452 if (!error || error == -ENOENT) {
1453 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1454 } else if (need_rehash)
1460 * To create a symbolic link, most file systems instantiate a new inode,
1461 * add a page to it containing the path, then write it out to the disk
1462 * using prepare_write/commit_write.
1464 * Unfortunately the NFS client can't create the in-core inode first
1465 * because it needs a file handle to create an in-core inode (see
1466 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1467 * symlink request has completed on the server.
1469 * So instead we allocate a raw page, copy the symname into it, then do
1470 * the SYMLINK request with the page as the buffer. If it succeeds, we
1471 * now have a new file handle and can instantiate an in-core NFS inode
1472 * and move the raw page into its mapping.
1474 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1476 struct pagevec lru_pvec;
1480 unsigned int pathlen = strlen(symname);
1483 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1484 dir->i_ino, dentry->d_name.name, symname);
1486 if (pathlen > PAGE_SIZE)
1487 return -ENAMETOOLONG;
1489 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1490 attr.ia_valid = ATTR_MODE;
1492 page = alloc_page(GFP_HIGHUSER);
1496 kaddr = kmap_atomic(page, KM_USER0);
1497 memcpy(kaddr, symname, pathlen);
1498 if (pathlen < PAGE_SIZE)
1499 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1500 kunmap_atomic(kaddr, KM_USER0);
1502 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1504 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1505 dir->i_sb->s_id, dir->i_ino,
1506 dentry->d_name.name, symname, error);
1513 * No big deal if we can't add this page to the page cache here.
1514 * READLINK will get the missing page from the server if needed.
1516 pagevec_init(&lru_pvec, 0);
1517 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1519 pagevec_add(&lru_pvec, page);
1520 pagevec_lru_add_file(&lru_pvec);
1521 SetPageUptodate(page);
1530 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1532 struct inode *inode = old_dentry->d_inode;
1535 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1536 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1537 dentry->d_parent->d_name.name, dentry->d_name.name);
1540 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1542 atomic_inc(&inode->i_count);
1543 d_add(dentry, inode);
1550 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1551 * different file handle for the same inode after a rename (e.g. when
1552 * moving to a different directory). A fail-safe method to do so would
1553 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1554 * rename the old file using the sillyrename stuff. This way, the original
1555 * file in old_dir will go away when the last process iput()s the inode.
1559 * It actually works quite well. One needs to have the possibility for
1560 * at least one ".nfs..." file in each directory the file ever gets
1561 * moved or linked to which happens automagically with the new
1562 * implementation that only depends on the dcache stuff instead of
1563 * using the inode layer
1565 * Unfortunately, things are a little more complicated than indicated
1566 * above. For a cross-directory move, we want to make sure we can get
1567 * rid of the old inode after the operation. This means there must be
1568 * no pending writes (if it's a file), and the use count must be 1.
1569 * If these conditions are met, we can drop the dentries before doing
1572 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1573 struct inode *new_dir, struct dentry *new_dentry)
1575 struct inode *old_inode = old_dentry->d_inode;
1576 struct inode *new_inode = new_dentry->d_inode;
1577 struct dentry *dentry = NULL, *rehash = NULL;
1581 * To prevent any new references to the target during the rename,
1582 * we unhash the dentry and free the inode in advance.
1584 if (!d_unhashed(new_dentry)) {
1586 rehash = new_dentry;
1589 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1590 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1591 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1592 atomic_read(&new_dentry->d_count));
1595 * First check whether the target is busy ... we can't
1596 * safely do _any_ rename if the target is in use.
1598 * For files, make a copy of the dentry and then do a
1599 * silly-rename. If the silly-rename succeeds, the
1600 * copied dentry is hashed and becomes the new target.
1604 if (S_ISDIR(new_inode->i_mode)) {
1606 if (!S_ISDIR(old_inode->i_mode))
1608 } else if (atomic_read(&new_dentry->d_count) > 2) {
1610 /* copy the target dentry's name */
1611 dentry = d_alloc(new_dentry->d_parent,
1612 &new_dentry->d_name);
1616 /* silly-rename the existing target ... */
1617 err = nfs_sillyrename(new_dir, new_dentry);
1619 new_dentry = rehash = dentry;
1621 /* instantiate the replacement target */
1622 d_instantiate(new_dentry, NULL);
1623 } else if (atomic_read(&new_dentry->d_count) > 1)
1624 /* dentry still busy? */
1627 nfs_drop_nlink(new_inode);
1631 * ... prune child dentries and writebacks if needed.
1633 if (atomic_read(&old_dentry->d_count) > 1) {
1634 if (S_ISREG(old_inode->i_mode))
1635 nfs_wb_all(old_inode);
1636 shrink_dcache_parent(old_dentry);
1638 nfs_inode_return_delegation(old_inode);
1640 if (new_inode != NULL) {
1641 nfs_inode_return_delegation(new_inode);
1642 d_delete(new_dentry);
1645 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1646 new_dir, &new_dentry->d_name);
1647 nfs_mark_for_revalidate(old_inode);
1652 d_move(old_dentry, new_dentry);
1653 nfs_set_verifier(new_dentry,
1654 nfs_save_change_attribute(new_dir));
1655 } else if (error == -ENOENT)
1656 nfs_dentry_handle_enoent(old_dentry);
1658 /* new dentry created? */
1664 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1665 static LIST_HEAD(nfs_access_lru_list);
1666 static atomic_long_t nfs_access_nr_entries;
1668 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1670 put_rpccred(entry->cred);
1672 smp_mb__before_atomic_dec();
1673 atomic_long_dec(&nfs_access_nr_entries);
1674 smp_mb__after_atomic_dec();
1677 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1680 struct nfs_inode *nfsi;
1681 struct nfs_access_entry *cache;
1684 spin_lock(&nfs_access_lru_lock);
1685 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1686 struct rw_semaphore *s_umount;
1687 struct inode *inode;
1689 if (nr_to_scan-- == 0)
1691 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1692 if (!down_read_trylock(s_umount))
1694 inode = igrab(&nfsi->vfs_inode);
1695 if (inode == NULL) {
1699 spin_lock(&inode->i_lock);
1700 if (list_empty(&nfsi->access_cache_entry_lru))
1701 goto remove_lru_entry;
1702 cache = list_entry(nfsi->access_cache_entry_lru.next,
1703 struct nfs_access_entry, lru);
1704 list_move(&cache->lru, &head);
1705 rb_erase(&cache->rb_node, &nfsi->access_cache);
1706 if (!list_empty(&nfsi->access_cache_entry_lru))
1707 list_move_tail(&nfsi->access_cache_inode_lru,
1708 &nfs_access_lru_list);
1711 list_del_init(&nfsi->access_cache_inode_lru);
1712 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1714 spin_unlock(&inode->i_lock);
1715 spin_unlock(&nfs_access_lru_lock);
1720 spin_unlock(&nfs_access_lru_lock);
1721 while (!list_empty(&head)) {
1722 cache = list_entry(head.next, struct nfs_access_entry, lru);
1723 list_del(&cache->lru);
1724 nfs_access_free_entry(cache);
1726 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1729 static void __nfs_access_zap_cache(struct inode *inode)
1731 struct nfs_inode *nfsi = NFS_I(inode);
1732 struct rb_root *root_node = &nfsi->access_cache;
1733 struct rb_node *n, *dispose = NULL;
1734 struct nfs_access_entry *entry;
1736 /* Unhook entries from the cache */
1737 while ((n = rb_first(root_node)) != NULL) {
1738 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1739 rb_erase(n, root_node);
1740 list_del(&entry->lru);
1741 n->rb_left = dispose;
1744 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1745 spin_unlock(&inode->i_lock);
1747 /* Now kill them all! */
1748 while (dispose != NULL) {
1750 dispose = n->rb_left;
1751 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1755 void nfs_access_zap_cache(struct inode *inode)
1757 /* Remove from global LRU init */
1758 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1759 spin_lock(&nfs_access_lru_lock);
1760 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1761 spin_unlock(&nfs_access_lru_lock);
1764 spin_lock(&inode->i_lock);
1765 /* This will release the spinlock */
1766 __nfs_access_zap_cache(inode);
1769 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1771 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1772 struct nfs_access_entry *entry;
1775 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1777 if (cred < entry->cred)
1779 else if (cred > entry->cred)
1787 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1789 struct nfs_inode *nfsi = NFS_I(inode);
1790 struct nfs_access_entry *cache;
1793 spin_lock(&inode->i_lock);
1794 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1796 cache = nfs_access_search_rbtree(inode, cred);
1799 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1801 res->jiffies = cache->jiffies;
1802 res->cred = cache->cred;
1803 res->mask = cache->mask;
1804 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1807 spin_unlock(&inode->i_lock);
1810 rb_erase(&cache->rb_node, &nfsi->access_cache);
1811 list_del(&cache->lru);
1812 spin_unlock(&inode->i_lock);
1813 nfs_access_free_entry(cache);
1816 /* This will release the spinlock */
1817 __nfs_access_zap_cache(inode);
1821 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1823 struct nfs_inode *nfsi = NFS_I(inode);
1824 struct rb_root *root_node = &nfsi->access_cache;
1825 struct rb_node **p = &root_node->rb_node;
1826 struct rb_node *parent = NULL;
1827 struct nfs_access_entry *entry;
1829 spin_lock(&inode->i_lock);
1830 while (*p != NULL) {
1832 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1834 if (set->cred < entry->cred)
1835 p = &parent->rb_left;
1836 else if (set->cred > entry->cred)
1837 p = &parent->rb_right;
1841 rb_link_node(&set->rb_node, parent, p);
1842 rb_insert_color(&set->rb_node, root_node);
1843 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1844 spin_unlock(&inode->i_lock);
1847 rb_replace_node(parent, &set->rb_node, root_node);
1848 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1849 list_del(&entry->lru);
1850 spin_unlock(&inode->i_lock);
1851 nfs_access_free_entry(entry);
1854 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1856 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1859 RB_CLEAR_NODE(&cache->rb_node);
1860 cache->jiffies = set->jiffies;
1861 cache->cred = get_rpccred(set->cred);
1862 cache->mask = set->mask;
1864 nfs_access_add_rbtree(inode, cache);
1866 /* Update accounting */
1867 smp_mb__before_atomic_inc();
1868 atomic_long_inc(&nfs_access_nr_entries);
1869 smp_mb__after_atomic_inc();
1871 /* Add inode to global LRU list */
1872 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
1873 spin_lock(&nfs_access_lru_lock);
1874 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1875 spin_unlock(&nfs_access_lru_lock);
1879 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1881 struct nfs_access_entry cache;
1884 status = nfs_access_get_cached(inode, cred, &cache);
1888 /* Be clever: ask server to check for all possible rights */
1889 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1891 cache.jiffies = jiffies;
1892 status = NFS_PROTO(inode)->access(inode, &cache);
1895 nfs_access_add_cache(inode, &cache);
1897 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1902 static int nfs_open_permission_mask(int openflags)
1906 if (openflags & FMODE_READ)
1908 if (openflags & FMODE_WRITE)
1910 if (openflags & FMODE_EXEC)
1915 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1917 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1920 int nfs_permission(struct inode *inode, int mask)
1922 struct rpc_cred *cred;
1925 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1927 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1929 /* Is this sys_access() ? */
1930 if (mask & MAY_ACCESS)
1933 switch (inode->i_mode & S_IFMT) {
1937 /* NFSv4 has atomic_open... */
1938 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1939 && (mask & MAY_OPEN))
1944 * Optimize away all write operations, since the server
1945 * will check permissions when we perform the op.
1947 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1952 if (!NFS_PROTO(inode)->access)
1955 cred = rpc_lookup_cred();
1956 if (!IS_ERR(cred)) {
1957 res = nfs_do_access(inode, cred, mask);
1960 res = PTR_ERR(cred);
1962 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1963 inode->i_sb->s_id, inode->i_ino, mask, res);
1966 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1968 res = generic_permission(inode, mask, NULL);
1974 * version-control: t
1975 * kept-new-versions: 5