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(VFS, "NFS: opendir(%s/%ld)\n",
137 inode->i_sb->s_id, inode->i_ino);
140 /* Call generic open code in order to cache credentials */
141 res = nfs_open(inode, filp);
146 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
150 unsigned long page_index;
153 loff_t current_index;
154 struct nfs_entry *entry;
155 decode_dirent_t decode;
158 unsigned long timestamp;
160 } nfs_readdir_descriptor_t;
162 /* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
175 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
177 struct file *file = desc->file;
178 struct inode *inode = file->f_path.dentry->d_inode;
179 struct rpc_cred *cred = nfs_file_cred(file);
180 unsigned long timestamp;
183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __FUNCTION__, (long long)desc->entry->cookie,
189 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
190 NFS_SERVER(inode)->dtsize, desc->plus);
192 /* We requested READDIRPLUS, but the server doesn't grok it */
193 if (error == -ENOTSUPP && desc->plus) {
194 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
195 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
201 desc->timestamp = timestamp;
202 desc->timestamp_valid = 1;
203 SetPageUptodate(page);
204 /* Ensure consistent page alignment of the data.
205 * Note: assumes we have exclusive access to this mapping either
206 * through inode->i_mutex or some other mechanism.
208 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
209 /* Should never happen */
210 nfs_zap_mapping(inode, inode->i_mapping);
221 int dir_decode(nfs_readdir_descriptor_t *desc)
223 __be32 *p = desc->ptr;
224 p = desc->decode(p, desc->entry, desc->plus);
228 if (desc->timestamp_valid)
229 desc->entry->fattr->time_start = desc->timestamp;
231 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
236 void dir_page_release(nfs_readdir_descriptor_t *desc)
239 page_cache_release(desc->page);
245 * Given a pointer to a buffer that has already been filled by a call
246 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
248 * If the end of the buffer has been reached, return -EAGAIN, if not,
249 * return the offset within the buffer of the next entry to be
253 int find_dirent(nfs_readdir_descriptor_t *desc)
255 struct nfs_entry *entry = desc->entry;
259 while((status = dir_decode(desc)) == 0) {
260 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
261 __FUNCTION__, (unsigned long long)entry->cookie);
262 if (entry->prev_cookie == *desc->dir_cookie)
264 if (loop_count++ > 200) {
273 * Given a pointer to a buffer that has already been filled by a call
274 * to readdir, find the entry at offset 'desc->file->f_pos'.
276 * If the end of the buffer has been reached, return -EAGAIN, if not,
277 * return the offset within the buffer of the next entry to be
281 int find_dirent_index(nfs_readdir_descriptor_t *desc)
283 struct nfs_entry *entry = desc->entry;
288 status = dir_decode(desc);
292 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
293 (unsigned long long)entry->cookie, desc->current_index);
295 if (desc->file->f_pos == desc->current_index) {
296 *desc->dir_cookie = entry->cookie;
299 desc->current_index++;
300 if (loop_count++ > 200) {
309 * Find the given page, and call find_dirent() or find_dirent_index in
310 * order to try to return the next entry.
313 int find_dirent_page(nfs_readdir_descriptor_t *desc)
315 struct inode *inode = desc->file->f_path.dentry->d_inode;
319 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
320 __FUNCTION__, desc->page_index,
321 (long long) *desc->dir_cookie);
323 /* If we find the page in the page_cache, we cannot be sure
324 * how fresh the data is, so we will ignore readdir_plus attributes.
326 desc->timestamp_valid = 0;
327 page = read_cache_page(inode->i_mapping, desc->page_index,
328 (filler_t *)nfs_readdir_filler, desc);
330 status = PTR_ERR(page);
334 /* NOTE: Someone else may have changed the READDIRPLUS flag */
336 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
337 if (*desc->dir_cookie != 0)
338 status = find_dirent(desc);
340 status = find_dirent_index(desc);
342 dir_page_release(desc);
344 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
349 * Recurse through the page cache pages, and return a
350 * filled nfs_entry structure of the next directory entry if possible.
352 * The target for the search is '*desc->dir_cookie' if non-0,
353 * 'desc->file->f_pos' otherwise
356 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
361 /* Always search-by-index from the beginning of the cache */
362 if (*desc->dir_cookie == 0) {
363 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
364 (long long)desc->file->f_pos);
365 desc->page_index = 0;
366 desc->entry->cookie = desc->entry->prev_cookie = 0;
367 desc->entry->eof = 0;
368 desc->current_index = 0;
370 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
371 (unsigned long long)*desc->dir_cookie);
374 res = find_dirent_page(desc);
377 /* Align to beginning of next page */
379 if (loop_count++ > 200) {
385 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
389 static inline unsigned int dt_type(struct inode *inode)
391 return (inode->i_mode >> 12) & 15;
394 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
397 * Once we've found the start of the dirent within a page: fill 'er up...
400 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
403 struct file *file = desc->file;
404 struct nfs_entry *entry = desc->entry;
405 struct dentry *dentry = NULL;
410 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
411 (unsigned long long)entry->cookie);
414 unsigned d_type = DT_UNKNOWN;
415 /* Note: entry->prev_cookie contains the cookie for
416 * retrieving the current dirent on the server */
419 /* Get a dentry if we have one */
422 dentry = nfs_readdir_lookup(desc);
424 /* Use readdirplus info */
425 if (dentry != NULL && dentry->d_inode != NULL) {
426 d_type = dt_type(dentry->d_inode);
427 fileid = NFS_FILEID(dentry->d_inode);
430 res = filldir(dirent, entry->name, entry->len,
431 file->f_pos, nfs_compat_user_ino64(fileid),
436 *desc->dir_cookie = entry->cookie;
437 if (dir_decode(desc) != 0) {
441 if (loop_count++ > 200) {
446 dir_page_release(desc);
449 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
450 (unsigned long long)*desc->dir_cookie, res);
455 * If we cannot find a cookie in our cache, we suspect that this is
456 * because it points to a deleted file, so we ask the server to return
457 * whatever it thinks is the next entry. We then feed this to filldir.
458 * If all goes well, we should then be able to find our way round the
459 * cache on the next call to readdir_search_pagecache();
461 * NOTE: we cannot add the anonymous page to the pagecache because
462 * the data it contains might not be page aligned. Besides,
463 * we should already have a complete representation of the
464 * directory in the page cache by the time we get here.
467 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
470 struct file *file = desc->file;
471 struct inode *inode = file->f_path.dentry->d_inode;
472 struct rpc_cred *cred = nfs_file_cred(file);
473 struct page *page = NULL;
475 unsigned long timestamp;
477 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
478 (unsigned long long)*desc->dir_cookie);
480 page = alloc_page(GFP_HIGHUSER);
486 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
488 NFS_SERVER(inode)->dtsize,
491 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
492 if (desc->error >= 0) {
493 desc->timestamp = timestamp;
494 desc->timestamp_valid = 1;
495 if ((status = dir_decode(desc)) == 0)
496 desc->entry->prev_cookie = *desc->dir_cookie;
502 status = nfs_do_filldir(desc, dirent, filldir);
504 /* Reset read descriptor so it searches the page cache from
505 * the start upon the next call to readdir_search_pagecache() */
506 desc->page_index = 0;
507 desc->entry->cookie = desc->entry->prev_cookie = 0;
508 desc->entry->eof = 0;
510 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
511 __FUNCTION__, status);
514 dir_page_release(desc);
518 /* The file offset position represents the dirent entry number. A
519 last cookie cache takes care of the common case of reading the
522 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
524 struct dentry *dentry = filp->f_path.dentry;
525 struct inode *inode = dentry->d_inode;
526 nfs_readdir_descriptor_t my_desc,
528 struct nfs_entry my_entry;
530 struct nfs_fattr fattr;
533 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
534 dentry->d_parent->d_name.name, dentry->d_name.name,
535 (long long)filp->f_pos);
536 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
541 * filp->f_pos points to the dirent entry number.
542 * *desc->dir_cookie has the cookie for the next entry. We have
543 * to either find the entry with the appropriate number or
544 * revalidate the cookie.
546 memset(desc, 0, sizeof(*desc));
549 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
550 desc->decode = NFS_PROTO(inode)->decode_dirent;
551 desc->plus = NFS_USE_READDIRPLUS(inode);
553 my_entry.cookie = my_entry.prev_cookie = 0;
556 my_entry.fattr = &fattr;
557 nfs_fattr_init(&fattr);
558 desc->entry = &my_entry;
560 nfs_block_sillyrename(dentry);
561 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
565 while(!desc->entry->eof) {
566 res = readdir_search_pagecache(desc);
568 if (res == -EBADCOOKIE) {
569 /* This means either end of directory */
570 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
571 /* Or that the server has 'lost' a cookie */
572 res = uncached_readdir(desc, dirent, filldir);
579 if (res == -ETOOSMALL && desc->plus) {
580 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
581 nfs_zap_caches(inode);
583 desc->entry->eof = 0;
589 res = nfs_do_filldir(desc, dirent, filldir);
596 nfs_unblock_sillyrename(dentry);
600 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
601 dentry->d_parent->d_name.name, dentry->d_name.name,
606 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
608 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
611 offset += filp->f_pos;
619 if (offset != filp->f_pos) {
620 filp->f_pos = offset;
621 nfs_file_open_context(filp)->dir_cookie = 0;
624 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
629 * All directory operations under NFS are synchronous, so fsync()
630 * is a dummy operation.
632 static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
634 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
635 dentry->d_parent->d_name.name, dentry->d_name.name,
642 * A check for whether or not the parent directory has changed.
643 * In the case it has, we assume that the dentries are untrustworthy
644 * and may need to be looked up again.
646 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
650 if (!nfs_verify_change_attribute(dir, dentry->d_time))
652 /* Revalidate nfsi->cache_change_attribute before we declare a match */
653 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
655 if (!nfs_verify_change_attribute(dir, dentry->d_time))
661 * Return the intent data that applies to this particular path component
663 * Note that the current set of intents only apply to the very last
664 * component of the path.
665 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
667 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
669 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
671 return nd->flags & mask;
675 * Use intent information to check whether or not we're going to do
676 * an O_EXCL create using this path component.
678 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
680 if (NFS_PROTO(dir)->version == 2)
682 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
684 return (nd->intent.open.flags & O_EXCL) != 0;
688 * Inode and filehandle revalidation for lookups.
690 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
691 * or if the intent information indicates that we're about to open this
692 * particular file and the "nocto" mount flag is not set.
696 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
698 struct nfs_server *server = NFS_SERVER(inode);
701 /* VFS wants an on-the-wire revalidation */
702 if (nd->flags & LOOKUP_REVAL)
704 /* This is an open(2) */
705 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
706 !(server->flags & NFS_MOUNT_NOCTO) &&
707 (S_ISREG(inode->i_mode) ||
708 S_ISDIR(inode->i_mode)))
712 return nfs_revalidate_inode(server, inode);
714 return __nfs_revalidate_inode(server, inode);
718 * We judge how long we want to trust negative
719 * dentries by looking at the parent inode mtime.
721 * If parent mtime has changed, we revalidate, else we wait for a
722 * period corresponding to the parent's attribute cache timeout value.
725 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
726 struct nameidata *nd)
728 /* Don't revalidate a negative dentry if we're creating a new file */
729 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
731 return !nfs_check_verifier(dir, dentry);
735 * This is called every time the dcache has a lookup hit,
736 * and we should check whether we can really trust that
739 * NOTE! The hit can be a negative hit too, don't assume
742 * If the parent directory is seen to have changed, we throw out the
743 * cached dentry and do a new lookup.
745 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
749 struct dentry *parent;
751 struct nfs_fh fhandle;
752 struct nfs_fattr fattr;
754 parent = dget_parent(dentry);
756 dir = parent->d_inode;
757 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
758 inode = dentry->d_inode;
761 if (nfs_neg_need_reval(dir, dentry, nd))
766 if (is_bad_inode(inode)) {
767 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
768 __FUNCTION__, dentry->d_parent->d_name.name,
769 dentry->d_name.name);
773 /* Force a full look up iff the parent directory has changed */
774 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
775 if (nfs_lookup_verify_inode(inode, nd))
780 if (NFS_STALE(inode))
783 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
786 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
788 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
791 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
795 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
796 __FUNCTION__, dentry->d_parent->d_name.name,
797 dentry->d_name.name);
802 nfs_mark_for_revalidate(dir);
803 if (inode && S_ISDIR(inode->i_mode)) {
804 /* Purge readdir caches. */
805 nfs_zap_caches(inode);
806 /* If we have submounts, don't unhash ! */
807 if (have_submounts(dentry))
809 shrink_dcache_parent(dentry);
814 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
815 __FUNCTION__, dentry->d_parent->d_name.name,
816 dentry->d_name.name);
821 * This is called from dput() when d_count is going to 0.
823 static int nfs_dentry_delete(struct dentry *dentry)
825 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
826 dentry->d_parent->d_name.name, dentry->d_name.name,
829 /* Unhash any dentry with a stale inode */
830 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
833 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
834 /* Unhash it, so that ->d_iput() would be called */
837 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
838 /* Unhash it, so that ancestors of killed async unlink
839 * files will be cleaned up during umount */
847 * Called when the dentry loses inode.
848 * We use it to clean up silly-renamed files.
850 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
852 nfs_inode_return_delegation(inode);
853 if (S_ISDIR(inode->i_mode))
854 /* drop any readdir cache as it could easily be old */
855 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
857 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
860 nfs_complete_unlink(dentry, inode);
866 struct dentry_operations nfs_dentry_operations = {
867 .d_revalidate = nfs_lookup_revalidate,
868 .d_delete = nfs_dentry_delete,
869 .d_iput = nfs_dentry_iput,
872 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
875 struct dentry *parent;
876 struct inode *inode = NULL;
878 struct nfs_fh fhandle;
879 struct nfs_fattr fattr;
881 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
882 dentry->d_parent->d_name.name, dentry->d_name.name);
883 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
885 res = ERR_PTR(-ENAMETOOLONG);
886 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
889 res = ERR_PTR(-ENOMEM);
890 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
895 * If we're doing an exclusive create, optimize away the lookup
896 * but don't hash the dentry.
898 if (nfs_is_exclusive_create(dir, nd)) {
899 d_instantiate(dentry, NULL);
904 parent = dentry->d_parent;
905 /* Protect against concurrent sillydeletes */
906 nfs_block_sillyrename(parent);
907 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
908 if (error == -ENOENT)
911 res = ERR_PTR(error);
912 goto out_unblock_sillyrename;
914 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
915 res = (struct dentry *)inode;
917 goto out_unblock_sillyrename;
920 res = d_materialise_unique(dentry, inode);
923 goto out_unblock_sillyrename;
926 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
927 out_unblock_sillyrename:
928 nfs_unblock_sillyrename(parent);
936 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
938 struct dentry_operations nfs4_dentry_operations = {
939 .d_revalidate = nfs_open_revalidate,
940 .d_delete = nfs_dentry_delete,
941 .d_iput = nfs_dentry_iput,
945 * Use intent information to determine whether we need to substitute
946 * the NFSv4-style stateful OPEN for the LOOKUP call
948 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
950 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
952 /* NFS does not (yet) have a stateful open for directories */
953 if (nd->flags & LOOKUP_DIRECTORY)
955 /* Are we trying to write to a read only partition? */
956 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
961 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
963 struct dentry *res = NULL;
966 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
967 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
969 /* Check that we are indeed trying to open this file */
970 if (!is_atomic_open(dir, nd))
973 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
974 res = ERR_PTR(-ENAMETOOLONG);
977 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
979 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
981 if (nd->intent.open.flags & O_EXCL) {
982 d_instantiate(dentry, NULL);
986 /* Open the file on the server */
988 res = nfs4_atomic_open(dir, dentry, nd);
991 error = PTR_ERR(res);
993 /* Make a negative dentry */
997 /* This turned out not to be a regular file */
1002 if (!(nd->intent.open.flags & O_NOFOLLOW))
1008 } else if (res != NULL)
1013 return nfs_lookup(dir, dentry, nd);
1016 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1018 struct dentry *parent = NULL;
1019 struct inode *inode = dentry->d_inode;
1021 int openflags, ret = 0;
1023 parent = dget_parent(dentry);
1024 dir = parent->d_inode;
1025 if (!is_atomic_open(dir, nd))
1027 /* We can't create new files in nfs_open_revalidate(), so we
1028 * optimize away revalidation of negative dentries.
1030 if (inode == NULL) {
1031 if (!nfs_neg_need_reval(dir, dentry, nd))
1036 /* NFS only supports OPEN on regular files */
1037 if (!S_ISREG(inode->i_mode))
1039 openflags = nd->intent.open.flags;
1040 /* We cannot do exclusive creation on a positive dentry */
1041 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1043 /* We can't create new files, or truncate existing ones here */
1044 openflags &= ~(O_CREAT|O_TRUNC);
1047 * Note: we're not holding inode->i_mutex and so may be racing with
1048 * operations that change the directory. We therefore save the
1049 * change attribute *before* we do the RPC call.
1052 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1061 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1063 return nfs_lookup_revalidate(dentry, nd);
1065 #endif /* CONFIG_NFSV4 */
1067 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1069 struct dentry *parent = desc->file->f_path.dentry;
1070 struct inode *dir = parent->d_inode;
1071 struct nfs_entry *entry = desc->entry;
1072 struct dentry *dentry, *alias;
1073 struct qstr name = {
1074 .name = entry->name,
1077 struct inode *inode;
1078 unsigned long verf = nfs_save_change_attribute(dir);
1082 if (name.name[0] == '.' && name.name[1] == '.')
1083 return dget_parent(parent);
1086 if (name.name[0] == '.')
1087 return dget(parent);
1090 spin_lock(&dir->i_lock);
1091 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1092 spin_unlock(&dir->i_lock);
1095 spin_unlock(&dir->i_lock);
1097 name.hash = full_name_hash(name.name, name.len);
1098 dentry = d_lookup(parent, &name);
1099 if (dentry != NULL) {
1100 /* Is this a positive dentry that matches the readdir info? */
1101 if (dentry->d_inode != NULL &&
1102 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1103 d_mountpoint(dentry))) {
1104 if (!desc->plus || entry->fh->size == 0)
1106 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1110 /* No, so d_drop to allow one to be created */
1114 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1116 if (name.len > NFS_SERVER(dir)->namelen)
1118 /* Note: caller is already holding the dir->i_mutex! */
1119 dentry = d_alloc(parent, &name);
1122 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1123 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1124 if (IS_ERR(inode)) {
1129 alias = d_materialise_unique(dentry, inode);
1130 if (alias != NULL) {
1138 nfs_set_verifier(dentry, verf);
1143 * Code common to create, mkdir, and mknod.
1145 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1146 struct nfs_fattr *fattr)
1148 struct dentry *parent = dget_parent(dentry);
1149 struct inode *dir = parent->d_inode;
1150 struct inode *inode;
1151 int error = -EACCES;
1155 /* We may have been initialized further down */
1156 if (dentry->d_inode)
1158 if (fhandle->size == 0) {
1159 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1163 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1164 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1165 struct nfs_server *server = NFS_SB(dentry->d_sb);
1166 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1170 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1171 error = PTR_ERR(inode);
1174 d_add(dentry, inode);
1179 nfs_mark_for_revalidate(dir);
1185 * Following a failed create operation, we drop the dentry rather
1186 * than retain a negative dentry. This avoids a problem in the event
1187 * that the operation succeeded on the server, but an error in the
1188 * reply path made it appear to have failed.
1190 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1191 struct nameidata *nd)
1197 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1198 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1200 attr.ia_mode = mode;
1201 attr.ia_valid = ATTR_MODE;
1203 if ((nd->flags & LOOKUP_CREATE) != 0)
1204 open_flags = nd->intent.open.flags;
1207 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1219 * See comments for nfs_proc_create regarding failed operations.
1222 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1227 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1228 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1230 if (!new_valid_dev(rdev))
1233 attr.ia_mode = mode;
1234 attr.ia_valid = ATTR_MODE;
1237 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1249 * See comments for nfs_proc_create regarding failed operations.
1251 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1256 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1257 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1259 attr.ia_valid = ATTR_MODE;
1260 attr.ia_mode = mode | S_IFDIR;
1263 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1274 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1276 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1280 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1284 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1285 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1288 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1289 /* Ensure the VFS deletes this inode */
1290 if (error == 0 && dentry->d_inode != NULL)
1291 clear_nlink(dentry->d_inode);
1292 else if (error == -ENOENT)
1293 nfs_dentry_handle_enoent(dentry);
1299 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1301 static unsigned int sillycounter;
1302 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
1303 const int countersize = sizeof(sillycounter)*2;
1304 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
1307 struct dentry *sdentry;
1310 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1311 dentry->d_parent->d_name.name, dentry->d_name.name,
1312 atomic_read(&dentry->d_count));
1313 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1316 * We don't allow a dentry to be silly-renamed twice.
1319 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1322 sprintf(silly, ".nfs%*.*Lx",
1323 fileidsize, fileidsize,
1324 (unsigned long long)NFS_FILEID(dentry->d_inode));
1326 /* Return delegation in anticipation of the rename */
1327 nfs_inode_return_delegation(dentry->d_inode);
1331 char *suffix = silly + slen - countersize;
1335 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1337 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1338 dentry->d_name.name, silly);
1340 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1342 * N.B. Better to return EBUSY here ... it could be
1343 * dangerous to delete the file while it's in use.
1345 if (IS_ERR(sdentry))
1347 } while(sdentry->d_inode != NULL); /* need negative lookup */
1349 qsilly.name = silly;
1350 qsilly.len = strlen(silly);
1351 if (dentry->d_inode) {
1352 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1354 nfs_mark_for_revalidate(dentry->d_inode);
1356 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1359 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1360 d_move(dentry, sdentry);
1361 error = nfs_async_unlink(dir, dentry);
1362 /* If we return 0 we don't unlink */
1370 * Remove a file after making sure there are no pending writes,
1371 * and after checking that the file has only one user.
1373 * We invalidate the attribute cache and free the inode prior to the operation
1374 * to avoid possible races if the server reuses the inode.
1376 static int nfs_safe_remove(struct dentry *dentry)
1378 struct inode *dir = dentry->d_parent->d_inode;
1379 struct inode *inode = dentry->d_inode;
1382 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1383 dentry->d_parent->d_name.name, dentry->d_name.name);
1385 /* If the dentry was sillyrenamed, we simply call d_delete() */
1386 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1391 if (inode != NULL) {
1392 nfs_inode_return_delegation(inode);
1393 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1394 /* The VFS may want to delete this inode */
1397 nfs_mark_for_revalidate(inode);
1399 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1400 if (error == -ENOENT)
1401 nfs_dentry_handle_enoent(dentry);
1406 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1407 * belongs to an active ".nfs..." file and we return -EBUSY.
1409 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1411 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1414 int need_rehash = 0;
1416 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1417 dir->i_ino, dentry->d_name.name);
1420 spin_lock(&dcache_lock);
1421 spin_lock(&dentry->d_lock);
1422 if (atomic_read(&dentry->d_count) > 1) {
1423 spin_unlock(&dentry->d_lock);
1424 spin_unlock(&dcache_lock);
1425 /* Start asynchronous writeout of the inode */
1426 write_inode_now(dentry->d_inode, 0);
1427 error = nfs_sillyrename(dir, dentry);
1431 if (!d_unhashed(dentry)) {
1435 spin_unlock(&dentry->d_lock);
1436 spin_unlock(&dcache_lock);
1437 error = nfs_safe_remove(dentry);
1438 if (!error || error == -ENOENT) {
1439 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1440 } else if (need_rehash)
1447 * To create a symbolic link, most file systems instantiate a new inode,
1448 * add a page to it containing the path, then write it out to the disk
1449 * using prepare_write/commit_write.
1451 * Unfortunately the NFS client can't create the in-core inode first
1452 * because it needs a file handle to create an in-core inode (see
1453 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1454 * symlink request has completed on the server.
1456 * So instead we allocate a raw page, copy the symname into it, then do
1457 * the SYMLINK request with the page as the buffer. If it succeeds, we
1458 * now have a new file handle and can instantiate an in-core NFS inode
1459 * and move the raw page into its mapping.
1461 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1463 struct pagevec lru_pvec;
1467 unsigned int pathlen = strlen(symname);
1470 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1471 dir->i_ino, dentry->d_name.name, symname);
1473 if (pathlen > PAGE_SIZE)
1474 return -ENAMETOOLONG;
1476 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1477 attr.ia_valid = ATTR_MODE;
1481 page = alloc_page(GFP_HIGHUSER);
1487 kaddr = kmap_atomic(page, KM_USER0);
1488 memcpy(kaddr, symname, pathlen);
1489 if (pathlen < PAGE_SIZE)
1490 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1491 kunmap_atomic(kaddr, KM_USER0);
1493 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1495 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1496 dir->i_sb->s_id, dir->i_ino,
1497 dentry->d_name.name, symname, error);
1505 * No big deal if we can't add this page to the page cache here.
1506 * READLINK will get the missing page from the server if needed.
1508 pagevec_init(&lru_pvec, 0);
1509 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1511 pagevec_add(&lru_pvec, page);
1512 pagevec_lru_add(&lru_pvec);
1513 SetPageUptodate(page);
1523 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1525 struct inode *inode = old_dentry->d_inode;
1528 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1529 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1530 dentry->d_parent->d_name.name, dentry->d_name.name);
1534 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1536 atomic_inc(&inode->i_count);
1537 d_add(dentry, inode);
1545 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1546 * different file handle for the same inode after a rename (e.g. when
1547 * moving to a different directory). A fail-safe method to do so would
1548 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1549 * rename the old file using the sillyrename stuff. This way, the original
1550 * file in old_dir will go away when the last process iput()s the inode.
1554 * It actually works quite well. One needs to have the possibility for
1555 * at least one ".nfs..." file in each directory the file ever gets
1556 * moved or linked to which happens automagically with the new
1557 * implementation that only depends on the dcache stuff instead of
1558 * using the inode layer
1560 * Unfortunately, things are a little more complicated than indicated
1561 * above. For a cross-directory move, we want to make sure we can get
1562 * rid of the old inode after the operation. This means there must be
1563 * no pending writes (if it's a file), and the use count must be 1.
1564 * If these conditions are met, we can drop the dentries before doing
1567 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1568 struct inode *new_dir, struct dentry *new_dentry)
1570 struct inode *old_inode = old_dentry->d_inode;
1571 struct inode *new_inode = new_dentry->d_inode;
1572 struct dentry *dentry = NULL, *rehash = NULL;
1576 * To prevent any new references to the target during the rename,
1577 * we unhash the dentry and free the inode in advance.
1580 if (!d_unhashed(new_dentry)) {
1582 rehash = new_dentry;
1585 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1586 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1587 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1588 atomic_read(&new_dentry->d_count));
1591 * First check whether the target is busy ... we can't
1592 * safely do _any_ rename if the target is in use.
1594 * For files, make a copy of the dentry and then do a
1595 * silly-rename. If the silly-rename succeeds, the
1596 * copied dentry is hashed and becomes the new target.
1600 if (S_ISDIR(new_inode->i_mode)) {
1602 if (!S_ISDIR(old_inode->i_mode))
1604 } else if (atomic_read(&new_dentry->d_count) > 2) {
1606 /* copy the target dentry's name */
1607 dentry = d_alloc(new_dentry->d_parent,
1608 &new_dentry->d_name);
1612 /* silly-rename the existing target ... */
1613 err = nfs_sillyrename(new_dir, new_dentry);
1615 new_dentry = rehash = dentry;
1617 /* instantiate the replacement target */
1618 d_instantiate(new_dentry, NULL);
1619 } else if (atomic_read(&new_dentry->d_count) > 1)
1620 /* dentry still busy? */
1623 drop_nlink(new_inode);
1627 * ... prune child dentries and writebacks if needed.
1629 if (atomic_read(&old_dentry->d_count) > 1) {
1630 if (S_ISREG(old_inode->i_mode))
1631 nfs_wb_all(old_inode);
1632 shrink_dcache_parent(old_dentry);
1634 nfs_inode_return_delegation(old_inode);
1636 if (new_inode != NULL) {
1637 nfs_inode_return_delegation(new_inode);
1638 d_delete(new_dentry);
1641 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1642 new_dir, &new_dentry->d_name);
1643 nfs_mark_for_revalidate(old_inode);
1648 d_move(old_dentry, new_dentry);
1649 nfs_set_verifier(new_dentry,
1650 nfs_save_change_attribute(new_dir));
1651 } else if (error == -ENOENT)
1652 nfs_dentry_handle_enoent(old_dentry);
1654 /* new dentry created? */
1661 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1662 static LIST_HEAD(nfs_access_lru_list);
1663 static atomic_long_t nfs_access_nr_entries;
1665 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1667 put_rpccred(entry->cred);
1669 smp_mb__before_atomic_dec();
1670 atomic_long_dec(&nfs_access_nr_entries);
1671 smp_mb__after_atomic_dec();
1674 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1677 struct nfs_inode *nfsi;
1678 struct nfs_access_entry *cache;
1681 spin_lock(&nfs_access_lru_lock);
1682 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1683 struct inode *inode;
1685 if (nr_to_scan-- == 0)
1687 inode = igrab(&nfsi->vfs_inode);
1690 spin_lock(&inode->i_lock);
1691 if (list_empty(&nfsi->access_cache_entry_lru))
1692 goto remove_lru_entry;
1693 cache = list_entry(nfsi->access_cache_entry_lru.next,
1694 struct nfs_access_entry, lru);
1695 list_move(&cache->lru, &head);
1696 rb_erase(&cache->rb_node, &nfsi->access_cache);
1697 if (!list_empty(&nfsi->access_cache_entry_lru))
1698 list_move_tail(&nfsi->access_cache_inode_lru,
1699 &nfs_access_lru_list);
1702 list_del_init(&nfsi->access_cache_inode_lru);
1703 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1705 spin_unlock(&inode->i_lock);
1706 spin_unlock(&nfs_access_lru_lock);
1710 spin_unlock(&nfs_access_lru_lock);
1711 while (!list_empty(&head)) {
1712 cache = list_entry(head.next, struct nfs_access_entry, lru);
1713 list_del(&cache->lru);
1714 nfs_access_free_entry(cache);
1716 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1719 static void __nfs_access_zap_cache(struct inode *inode)
1721 struct nfs_inode *nfsi = NFS_I(inode);
1722 struct rb_root *root_node = &nfsi->access_cache;
1723 struct rb_node *n, *dispose = NULL;
1724 struct nfs_access_entry *entry;
1726 /* Unhook entries from the cache */
1727 while ((n = rb_first(root_node)) != NULL) {
1728 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1729 rb_erase(n, root_node);
1730 list_del(&entry->lru);
1731 n->rb_left = dispose;
1734 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1735 spin_unlock(&inode->i_lock);
1737 /* Now kill them all! */
1738 while (dispose != NULL) {
1740 dispose = n->rb_left;
1741 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1745 void nfs_access_zap_cache(struct inode *inode)
1747 /* Remove from global LRU init */
1748 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1749 spin_lock(&nfs_access_lru_lock);
1750 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1751 spin_unlock(&nfs_access_lru_lock);
1754 spin_lock(&inode->i_lock);
1755 /* This will release the spinlock */
1756 __nfs_access_zap_cache(inode);
1759 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1761 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1762 struct nfs_access_entry *entry;
1765 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1767 if (cred < entry->cred)
1769 else if (cred > entry->cred)
1777 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1779 struct nfs_inode *nfsi = NFS_I(inode);
1780 struct nfs_access_entry *cache;
1783 spin_lock(&inode->i_lock);
1784 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1786 cache = nfs_access_search_rbtree(inode, cred);
1789 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
1791 res->jiffies = cache->jiffies;
1792 res->cred = cache->cred;
1793 res->mask = cache->mask;
1794 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1797 spin_unlock(&inode->i_lock);
1800 rb_erase(&cache->rb_node, &nfsi->access_cache);
1801 list_del(&cache->lru);
1802 spin_unlock(&inode->i_lock);
1803 nfs_access_free_entry(cache);
1806 /* This will release the spinlock */
1807 __nfs_access_zap_cache(inode);
1811 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1813 struct nfs_inode *nfsi = NFS_I(inode);
1814 struct rb_root *root_node = &nfsi->access_cache;
1815 struct rb_node **p = &root_node->rb_node;
1816 struct rb_node *parent = NULL;
1817 struct nfs_access_entry *entry;
1819 spin_lock(&inode->i_lock);
1820 while (*p != NULL) {
1822 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1824 if (set->cred < entry->cred)
1825 p = &parent->rb_left;
1826 else if (set->cred > entry->cred)
1827 p = &parent->rb_right;
1831 rb_link_node(&set->rb_node, parent, p);
1832 rb_insert_color(&set->rb_node, root_node);
1833 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1834 spin_unlock(&inode->i_lock);
1837 rb_replace_node(parent, &set->rb_node, root_node);
1838 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1839 list_del(&entry->lru);
1840 spin_unlock(&inode->i_lock);
1841 nfs_access_free_entry(entry);
1844 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1846 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1849 RB_CLEAR_NODE(&cache->rb_node);
1850 cache->jiffies = set->jiffies;
1851 cache->cred = get_rpccred(set->cred);
1852 cache->mask = set->mask;
1854 nfs_access_add_rbtree(inode, cache);
1856 /* Update accounting */
1857 smp_mb__before_atomic_inc();
1858 atomic_long_inc(&nfs_access_nr_entries);
1859 smp_mb__after_atomic_inc();
1861 /* Add inode to global LRU list */
1862 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1863 spin_lock(&nfs_access_lru_lock);
1864 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1865 spin_unlock(&nfs_access_lru_lock);
1869 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1871 struct nfs_access_entry cache;
1874 status = nfs_access_get_cached(inode, cred, &cache);
1878 /* Be clever: ask server to check for all possible rights */
1879 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1881 cache.jiffies = jiffies;
1882 status = NFS_PROTO(inode)->access(inode, &cache);
1885 nfs_access_add_cache(inode, &cache);
1887 if ((cache.mask & mask) == mask)
1892 static int nfs_open_permission_mask(int openflags)
1896 if (openflags & FMODE_READ)
1898 if (openflags & FMODE_WRITE)
1900 if (openflags & FMODE_EXEC)
1905 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1907 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1910 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1912 struct rpc_cred *cred;
1915 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1919 /* Is this sys_access() ? */
1920 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1923 switch (inode->i_mode & S_IFMT) {
1927 /* NFSv4 has atomic_open... */
1928 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1930 && (nd->flags & LOOKUP_OPEN))
1935 * Optimize away all write operations, since the server
1936 * will check permissions when we perform the op.
1938 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1945 if (!NFS_PROTO(inode)->access)
1948 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1949 if (!IS_ERR(cred)) {
1950 res = nfs_do_access(inode, cred, mask);
1953 res = PTR_ERR(cred);
1956 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1957 inode->i_sb->s_id, inode->i_ino, mask, res);
1960 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1962 res = generic_permission(inode, mask, NULL);
1969 * version-control: t
1970 * kept-new-versions: 5