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/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_readdir(struct file *, void *, filldir_t);
48 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
49 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
50 static int nfs_mkdir(struct inode *, struct dentry *, int);
51 static int nfs_rmdir(struct inode *, struct dentry *);
52 static int nfs_unlink(struct inode *, struct dentry *);
53 static int nfs_symlink(struct inode *, struct dentry *, const char *);
54 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
55 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
56 static int nfs_rename(struct inode *, struct dentry *,
57 struct inode *, struct dentry *);
58 static int nfs_fsync_dir(struct file *, int);
59 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 static void nfs_readdir_clear_array(struct page*);
62 const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .readdir = nfs_readdir,
67 .release = nfs_release,
68 .fsync = nfs_fsync_dir,
71 const struct inode_operations nfs_dir_inode_operations = {
76 .symlink = nfs_symlink,
81 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
86 const struct address_space_operations nfs_dir_aops = {
87 .freepage = nfs_readdir_clear_array,
91 const struct inode_operations nfs3_dir_inode_operations = {
96 .symlink = nfs_symlink,
100 .rename = nfs_rename,
101 .permission = nfs_permission,
102 .getattr = nfs_getattr,
103 .setattr = nfs_setattr,
104 .listxattr = nfs3_listxattr,
105 .getxattr = nfs3_getxattr,
106 .setxattr = nfs3_setxattr,
107 .removexattr = nfs3_removexattr,
109 #endif /* CONFIG_NFS_V3 */
113 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
114 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
115 const struct inode_operations nfs4_dir_inode_operations = {
116 .create = nfs_open_create,
117 .lookup = nfs_atomic_lookup,
119 .unlink = nfs_unlink,
120 .symlink = nfs_symlink,
124 .rename = nfs_rename,
125 .permission = nfs_permission,
126 .getattr = nfs_getattr,
127 .setattr = nfs_setattr,
128 .getxattr = generic_getxattr,
129 .setxattr = generic_setxattr,
130 .listxattr = generic_listxattr,
131 .removexattr = generic_removexattr,
134 #endif /* CONFIG_NFS_V4 */
140 nfs_opendir(struct inode *inode, struct file *filp)
144 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
145 filp->f_path.dentry->d_parent->d_name.name,
146 filp->f_path.dentry->d_name.name);
148 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
150 /* Call generic open code in order to cache credentials */
151 res = nfs_open(inode, filp);
152 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
153 /* This is a mountpoint, so d_revalidate will never
154 * have been called, so we need to refresh the
155 * inode (for close-open consistency) ourselves.
157 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
162 struct nfs_cache_array_entry {
166 unsigned char d_type;
169 struct nfs_cache_array {
173 struct nfs_cache_array_entry array[0];
176 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
180 unsigned long page_index;
183 loff_t current_index;
184 decode_dirent_t decode;
186 unsigned long timestamp;
187 unsigned long gencount;
188 unsigned int cache_entry_index;
191 } nfs_readdir_descriptor_t;
194 * The caller is responsible for calling nfs_readdir_release_array(page)
197 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
201 return ERR_PTR(-EIO);
204 return ERR_PTR(-ENOMEM);
209 void nfs_readdir_release_array(struct page *page)
215 * we are freeing strings created by nfs_add_to_readdir_array()
218 void nfs_readdir_clear_array(struct page *page)
220 struct nfs_cache_array *array;
223 array = kmap_atomic(page, KM_USER0);
224 for (i = 0; i < array->size; i++)
225 kfree(array->array[i].string.name);
226 kunmap_atomic(array, KM_USER0);
230 * the caller is responsible for freeing qstr.name
231 * when called by nfs_readdir_add_to_array, the strings will be freed in
232 * nfs_clear_readdir_array()
235 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
238 string->name = kmemdup(name, len, GFP_KERNEL);
239 if (string->name == NULL)
242 * Avoid a kmemleak false positive. The pointer to the name is stored
243 * in a page cache page which kmemleak does not scan.
245 kmemleak_not_leak(string->name);
246 string->hash = full_name_hash(name, len);
251 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
253 struct nfs_cache_array *array = nfs_readdir_get_array(page);
254 struct nfs_cache_array_entry *cache_entry;
258 return PTR_ERR(array);
260 cache_entry = &array->array[array->size];
262 /* Check that this entry lies within the page bounds */
264 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
267 cache_entry->cookie = entry->prev_cookie;
268 cache_entry->ino = entry->ino;
269 cache_entry->d_type = entry->d_type;
270 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
273 array->last_cookie = entry->cookie;
276 array->eof_index = array->size;
278 nfs_readdir_release_array(page);
283 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
285 loff_t diff = desc->file->f_pos - desc->current_index;
290 if (diff >= array->size) {
291 if (array->eof_index >= 0)
293 desc->current_index += array->size;
297 index = (unsigned int)diff;
298 *desc->dir_cookie = array->array[index].cookie;
299 desc->cache_entry_index = index;
307 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
310 int status = -EAGAIN;
312 for (i = 0; i < array->size; i++) {
313 if (array->array[i].cookie == *desc->dir_cookie) {
314 desc->cache_entry_index = i;
318 if (array->eof_index >= 0) {
319 status = -EBADCOOKIE;
320 if (*desc->dir_cookie == array->last_cookie)
327 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
329 struct nfs_cache_array *array;
332 array = nfs_readdir_get_array(desc->page);
334 status = PTR_ERR(array);
338 if (*desc->dir_cookie == 0)
339 status = nfs_readdir_search_for_pos(array, desc);
341 status = nfs_readdir_search_for_cookie(array, desc);
343 if (status == -EAGAIN) {
344 desc->last_cookie = array->last_cookie;
347 nfs_readdir_release_array(desc->page);
352 /* Fill a page with xdr information before transferring to the cache page */
354 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
355 struct nfs_entry *entry, struct file *file, struct inode *inode)
357 struct rpc_cred *cred = nfs_file_cred(file);
358 unsigned long timestamp, gencount;
363 gencount = nfs_inc_attr_generation_counter();
364 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
365 NFS_SERVER(inode)->dtsize, desc->plus);
367 /* We requested READDIRPLUS, but the server doesn't grok it */
368 if (error == -ENOTSUPP && desc->plus) {
369 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
370 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
376 desc->timestamp = timestamp;
377 desc->gencount = gencount;
382 static int xdr_decode(nfs_readdir_descriptor_t *desc,
383 struct nfs_entry *entry, struct xdr_stream *xdr)
387 error = desc->decode(xdr, entry, desc->plus);
390 entry->fattr->time_start = desc->timestamp;
391 entry->fattr->gencount = desc->gencount;
396 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
398 if (dentry->d_inode == NULL)
400 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
408 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
410 struct qstr filename = {
414 struct dentry *dentry;
415 struct dentry *alias;
416 struct inode *dir = parent->d_inode;
419 if (filename.name[0] == '.') {
420 if (filename.len == 1)
422 if (filename.len == 2 && filename.name[1] == '.')
425 filename.hash = full_name_hash(filename.name, filename.len);
427 dentry = d_lookup(parent, &filename);
428 if (dentry != NULL) {
429 if (nfs_same_file(dentry, entry)) {
430 nfs_refresh_inode(dentry->d_inode, entry->fattr);
438 dentry = d_alloc(parent, &filename);
442 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
446 alias = d_materialise_unique(dentry, inode);
450 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
453 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
459 /* Perform conversion from xdr to cache array */
461 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
462 struct page **xdr_pages, struct page *page, unsigned int buflen)
464 struct xdr_stream stream;
465 struct xdr_buf buf = {
471 struct page *scratch;
472 struct nfs_cache_array *array;
473 unsigned int count = 0;
476 scratch = alloc_page(GFP_KERNEL);
480 xdr_init_decode(&stream, &buf, NULL);
481 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
484 status = xdr_decode(desc, entry, &stream);
486 if (status == -EAGAIN)
494 nfs_prime_dcache(desc->file->f_path.dentry, entry);
496 status = nfs_readdir_add_to_array(entry, page);
499 } while (!entry->eof);
501 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
502 array = nfs_readdir_get_array(page);
503 if (!IS_ERR(array)) {
504 array->eof_index = array->size;
506 nfs_readdir_release_array(page);
508 status = PTR_ERR(array);
516 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
519 for (i = 0; i < npages; i++)
524 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
527 nfs_readdir_free_pagearray(pages, npages);
531 * nfs_readdir_large_page will allocate pages that must be freed with a call
532 * to nfs_readdir_free_large_page
535 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
539 for (i = 0; i < npages; i++) {
540 struct page *page = alloc_page(GFP_KERNEL);
548 nfs_readdir_free_pagearray(pages, i);
553 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
555 struct page *pages[NFS_MAX_READDIR_PAGES];
556 void *pages_ptr = NULL;
557 struct nfs_entry entry;
558 struct file *file = desc->file;
559 struct nfs_cache_array *array;
560 int status = -ENOMEM;
561 unsigned int array_size = ARRAY_SIZE(pages);
563 entry.prev_cookie = 0;
564 entry.cookie = desc->last_cookie;
566 entry.fh = nfs_alloc_fhandle();
567 entry.fattr = nfs_alloc_fattr();
568 entry.server = NFS_SERVER(inode);
569 if (entry.fh == NULL || entry.fattr == NULL)
572 array = nfs_readdir_get_array(page);
574 status = PTR_ERR(array);
577 memset(array, 0, sizeof(struct nfs_cache_array));
578 array->eof_index = -1;
580 status = nfs_readdir_large_page(pages, array_size);
582 goto out_release_array;
585 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
590 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
592 if (status == -ENOSPC)
596 } while (array->eof_index < 0);
598 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
600 nfs_readdir_release_array(page);
602 nfs_free_fattr(entry.fattr);
603 nfs_free_fhandle(entry.fh);
608 * Now we cache directories properly, by converting xdr information
609 * to an array that can be used for lookups later. This results in
610 * fewer cache pages, since we can store more information on each page.
611 * We only need to convert from xdr once so future lookups are much simpler
614 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
616 struct inode *inode = desc->file->f_path.dentry->d_inode;
619 ret = nfs_readdir_xdr_to_array(desc, page, inode);
622 SetPageUptodate(page);
624 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
625 /* Should never happen */
626 nfs_zap_mapping(inode, inode->i_mapping);
636 void cache_page_release(nfs_readdir_descriptor_t *desc)
638 if (!desc->page->mapping)
639 nfs_readdir_clear_array(desc->page);
640 page_cache_release(desc->page);
645 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
647 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
648 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
652 * Returns 0 if desc->dir_cookie was found on page desc->page_index
655 int find_cache_page(nfs_readdir_descriptor_t *desc)
659 desc->page = get_cache_page(desc);
660 if (IS_ERR(desc->page))
661 return PTR_ERR(desc->page);
663 res = nfs_readdir_search_array(desc);
665 cache_page_release(desc);
669 /* Search for desc->dir_cookie from the beginning of the page cache */
671 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
675 if (desc->page_index == 0) {
676 desc->current_index = 0;
677 desc->last_cookie = 0;
680 res = find_cache_page(desc);
681 } while (res == -EAGAIN);
686 * Once we've found the start of the dirent within a page: fill 'er up...
689 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
692 struct file *file = desc->file;
695 struct nfs_cache_array *array = NULL;
697 array = nfs_readdir_get_array(desc->page);
699 res = PTR_ERR(array);
703 for (i = desc->cache_entry_index; i < array->size; i++) {
704 struct nfs_cache_array_entry *ent;
706 ent = &array->array[i];
707 if (filldir(dirent, ent->string.name, ent->string.len,
708 file->f_pos, nfs_compat_user_ino64(ent->ino),
714 if (i < (array->size-1))
715 *desc->dir_cookie = array->array[i+1].cookie;
717 *desc->dir_cookie = array->last_cookie;
719 if (array->eof_index >= 0)
722 nfs_readdir_release_array(desc->page);
724 cache_page_release(desc);
725 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
726 (unsigned long long)*desc->dir_cookie, res);
731 * If we cannot find a cookie in our cache, we suspect that this is
732 * because it points to a deleted file, so we ask the server to return
733 * whatever it thinks is the next entry. We then feed this to filldir.
734 * If all goes well, we should then be able to find our way round the
735 * cache on the next call to readdir_search_pagecache();
737 * NOTE: we cannot add the anonymous page to the pagecache because
738 * the data it contains might not be page aligned. Besides,
739 * we should already have a complete representation of the
740 * directory in the page cache by the time we get here.
743 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
746 struct page *page = NULL;
748 struct inode *inode = desc->file->f_path.dentry->d_inode;
750 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
751 (unsigned long long)*desc->dir_cookie);
753 page = alloc_page(GFP_HIGHUSER);
759 desc->page_index = 0;
760 desc->last_cookie = *desc->dir_cookie;
763 status = nfs_readdir_xdr_to_array(desc, page, inode);
767 status = nfs_do_filldir(desc, dirent, filldir);
770 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
774 cache_page_release(desc);
778 /* The file offset position represents the dirent entry number. A
779 last cookie cache takes care of the common case of reading the
782 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
784 struct dentry *dentry = filp->f_path.dentry;
785 struct inode *inode = dentry->d_inode;
786 nfs_readdir_descriptor_t my_desc,
790 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
791 dentry->d_parent->d_name.name, dentry->d_name.name,
792 (long long)filp->f_pos);
793 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
796 * filp->f_pos points to the dirent entry number.
797 * *desc->dir_cookie has the cookie for the next entry. We have
798 * to either find the entry with the appropriate number or
799 * revalidate the cookie.
801 memset(desc, 0, sizeof(*desc));
804 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
805 desc->decode = NFS_PROTO(inode)->decode_dirent;
806 desc->plus = NFS_USE_READDIRPLUS(inode);
808 nfs_block_sillyrename(dentry);
809 res = nfs_revalidate_mapping(inode, filp->f_mapping);
814 res = readdir_search_pagecache(desc);
816 if (res == -EBADCOOKIE) {
818 /* This means either end of directory */
819 if (*desc->dir_cookie && desc->eof == 0) {
820 /* Or that the server has 'lost' a cookie */
821 res = uncached_readdir(desc, dirent, filldir);
827 if (res == -ETOOSMALL && desc->plus) {
828 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
829 nfs_zap_caches(inode);
830 desc->page_index = 0;
838 res = nfs_do_filldir(desc, dirent, filldir);
841 } while (!desc->eof);
843 nfs_unblock_sillyrename(dentry);
846 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
847 dentry->d_parent->d_name.name, dentry->d_name.name,
852 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
854 struct dentry *dentry = filp->f_path.dentry;
855 struct inode *inode = dentry->d_inode;
857 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
858 dentry->d_parent->d_name.name,
862 mutex_lock(&inode->i_mutex);
865 offset += filp->f_pos;
873 if (offset != filp->f_pos) {
874 filp->f_pos = offset;
875 nfs_file_open_context(filp)->dir_cookie = 0;
878 mutex_unlock(&inode->i_mutex);
883 * All directory operations under NFS are synchronous, so fsync()
884 * is a dummy operation.
886 static int nfs_fsync_dir(struct file *filp, int datasync)
888 struct dentry *dentry = filp->f_path.dentry;
890 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
891 dentry->d_parent->d_name.name, dentry->d_name.name,
894 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
899 * nfs_force_lookup_revalidate - Mark the directory as having changed
900 * @dir - pointer to directory inode
902 * This forces the revalidation code in nfs_lookup_revalidate() to do a
903 * full lookup on all child dentries of 'dir' whenever a change occurs
904 * on the server that might have invalidated our dcache.
906 * The caller should be holding dir->i_lock
908 void nfs_force_lookup_revalidate(struct inode *dir)
910 NFS_I(dir)->cache_change_attribute++;
914 * A check for whether or not the parent directory has changed.
915 * In the case it has, we assume that the dentries are untrustworthy
916 * and may need to be looked up again.
918 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
922 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
924 if (!nfs_verify_change_attribute(dir, dentry->d_time))
926 /* Revalidate nfsi->cache_change_attribute before we declare a match */
927 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
929 if (!nfs_verify_change_attribute(dir, dentry->d_time))
935 * Return the intent data that applies to this particular path component
937 * Note that the current set of intents only apply to the very last
938 * component of the path.
939 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
941 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
944 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
946 return nd->flags & mask;
950 * Use intent information to check whether or not we're going to do
951 * an O_EXCL create using this path component.
953 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
955 if (NFS_PROTO(dir)->version == 2)
957 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
961 * Inode and filehandle revalidation for lookups.
963 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
964 * or if the intent information indicates that we're about to open this
965 * particular file and the "nocto" mount flag is not set.
969 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
971 struct nfs_server *server = NFS_SERVER(inode);
973 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
976 /* VFS wants an on-the-wire revalidation */
977 if (nd->flags & LOOKUP_REVAL)
979 /* This is an open(2) */
980 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
981 !(server->flags & NFS_MOUNT_NOCTO) &&
982 (S_ISREG(inode->i_mode) ||
983 S_ISDIR(inode->i_mode)))
987 return nfs_revalidate_inode(server, inode);
989 return __nfs_revalidate_inode(server, inode);
993 * We judge how long we want to trust negative
994 * dentries by looking at the parent inode mtime.
996 * If parent mtime has changed, we revalidate, else we wait for a
997 * period corresponding to the parent's attribute cache timeout value.
1000 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1001 struct nameidata *nd)
1003 /* Don't revalidate a negative dentry if we're creating a new file */
1004 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1006 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1008 return !nfs_check_verifier(dir, dentry);
1012 * This is called every time the dcache has a lookup hit,
1013 * and we should check whether we can really trust that
1016 * NOTE! The hit can be a negative hit too, don't assume
1019 * If the parent directory is seen to have changed, we throw out the
1020 * cached dentry and do a new lookup.
1022 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1025 struct inode *inode;
1026 struct dentry *parent;
1027 struct nfs_fh *fhandle = NULL;
1028 struct nfs_fattr *fattr = NULL;
1031 if (nd->flags & LOOKUP_RCU)
1034 parent = dget_parent(dentry);
1035 dir = parent->d_inode;
1036 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1037 inode = dentry->d_inode;
1040 if (nfs_neg_need_reval(dir, dentry, nd))
1045 if (is_bad_inode(inode)) {
1046 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1047 __func__, dentry->d_parent->d_name.name,
1048 dentry->d_name.name);
1052 if (nfs_have_delegation(inode, FMODE_READ))
1053 goto out_set_verifier;
1055 /* Force a full look up iff the parent directory has changed */
1056 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1057 if (nfs_lookup_verify_inode(inode, nd))
1058 goto out_zap_parent;
1062 if (NFS_STALE(inode))
1066 fhandle = nfs_alloc_fhandle();
1067 fattr = nfs_alloc_fattr();
1068 if (fhandle == NULL || fattr == NULL)
1071 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1074 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1076 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1079 nfs_free_fattr(fattr);
1080 nfs_free_fhandle(fhandle);
1082 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1085 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1086 __func__, dentry->d_parent->d_name.name,
1087 dentry->d_name.name);
1090 nfs_zap_caches(dir);
1092 nfs_mark_for_revalidate(dir);
1093 if (inode && S_ISDIR(inode->i_mode)) {
1094 /* Purge readdir caches. */
1095 nfs_zap_caches(inode);
1096 /* If we have submounts, don't unhash ! */
1097 if (have_submounts(dentry))
1099 if (dentry->d_flags & DCACHE_DISCONNECTED)
1101 shrink_dcache_parent(dentry);
1104 nfs_free_fattr(fattr);
1105 nfs_free_fhandle(fhandle);
1107 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1108 __func__, dentry->d_parent->d_name.name,
1109 dentry->d_name.name);
1112 nfs_free_fattr(fattr);
1113 nfs_free_fhandle(fhandle);
1115 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1116 __func__, dentry->d_parent->d_name.name,
1117 dentry->d_name.name, error);
1122 * This is called from dput() when d_count is going to 0.
1124 static int nfs_dentry_delete(const struct dentry *dentry)
1126 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1127 dentry->d_parent->d_name.name, dentry->d_name.name,
1130 /* Unhash any dentry with a stale inode */
1131 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1134 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1135 /* Unhash it, so that ->d_iput() would be called */
1138 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1139 /* Unhash it, so that ancestors of killed async unlink
1140 * files will be cleaned up during umount */
1147 static void nfs_drop_nlink(struct inode *inode)
1149 spin_lock(&inode->i_lock);
1150 if (inode->i_nlink > 0)
1152 spin_unlock(&inode->i_lock);
1156 * Called when the dentry loses inode.
1157 * We use it to clean up silly-renamed files.
1159 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1161 if (S_ISDIR(inode->i_mode))
1162 /* drop any readdir cache as it could easily be old */
1163 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1165 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1167 nfs_complete_unlink(dentry, inode);
1172 const struct dentry_operations nfs_dentry_operations = {
1173 .d_revalidate = nfs_lookup_revalidate,
1174 .d_delete = nfs_dentry_delete,
1175 .d_iput = nfs_dentry_iput,
1178 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1181 struct dentry *parent;
1182 struct inode *inode = NULL;
1183 struct nfs_fh *fhandle = NULL;
1184 struct nfs_fattr *fattr = NULL;
1187 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1188 dentry->d_parent->d_name.name, dentry->d_name.name);
1189 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1191 res = ERR_PTR(-ENAMETOOLONG);
1192 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1196 * If we're doing an exclusive create, optimize away the lookup
1197 * but don't hash the dentry.
1199 if (nfs_is_exclusive_create(dir, nd)) {
1200 d_instantiate(dentry, NULL);
1205 res = ERR_PTR(-ENOMEM);
1206 fhandle = nfs_alloc_fhandle();
1207 fattr = nfs_alloc_fattr();
1208 if (fhandle == NULL || fattr == NULL)
1211 parent = dentry->d_parent;
1212 /* Protect against concurrent sillydeletes */
1213 nfs_block_sillyrename(parent);
1214 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1215 if (error == -ENOENT)
1218 res = ERR_PTR(error);
1219 goto out_unblock_sillyrename;
1221 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1222 res = ERR_CAST(inode);
1224 goto out_unblock_sillyrename;
1227 res = d_materialise_unique(dentry, inode);
1230 goto out_unblock_sillyrename;
1233 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1234 out_unblock_sillyrename:
1235 nfs_unblock_sillyrename(parent);
1237 nfs_free_fattr(fattr);
1238 nfs_free_fhandle(fhandle);
1242 #ifdef CONFIG_NFS_V4
1243 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1245 const struct dentry_operations nfs4_dentry_operations = {
1246 .d_revalidate = nfs_open_revalidate,
1247 .d_delete = nfs_dentry_delete,
1248 .d_iput = nfs_dentry_iput,
1252 * Use intent information to determine whether we need to substitute
1253 * the NFSv4-style stateful OPEN for the LOOKUP call
1255 static int is_atomic_open(struct nameidata *nd)
1257 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1259 /* NFS does not (yet) have a stateful open for directories */
1260 if (nd->flags & LOOKUP_DIRECTORY)
1262 /* Are we trying to write to a read only partition? */
1263 if (__mnt_is_readonly(nd->path.mnt) &&
1264 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1269 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1271 struct path path = {
1272 .mnt = nd->path.mnt,
1275 struct nfs_open_context *ctx;
1276 struct rpc_cred *cred;
1277 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1279 cred = rpc_lookup_cred();
1281 return ERR_CAST(cred);
1282 ctx = alloc_nfs_open_context(&path, cred, fmode);
1285 return ERR_PTR(-ENOMEM);
1289 static int do_open(struct inode *inode, struct file *filp)
1291 nfs_fscache_set_inode_cookie(inode, filp);
1295 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1300 /* If the open_intent is for execute, we have an extra check to make */
1301 if (ctx->mode & FMODE_EXEC) {
1302 ret = nfs_may_open(ctx->path.dentry->d_inode,
1304 nd->intent.open.flags);
1308 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1310 ret = PTR_ERR(filp);
1312 nfs_file_set_open_context(filp, ctx);
1314 put_nfs_open_context(ctx);
1318 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1320 struct nfs_open_context *ctx;
1322 struct dentry *res = NULL;
1323 struct inode *inode;
1327 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1328 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1330 /* Check that we are indeed trying to open this file */
1331 if (!is_atomic_open(nd))
1334 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1335 res = ERR_PTR(-ENAMETOOLONG);
1339 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1341 if (nd->flags & LOOKUP_EXCL) {
1342 d_instantiate(dentry, NULL);
1346 ctx = nameidata_to_nfs_open_context(dentry, nd);
1347 res = ERR_CAST(ctx);
1351 open_flags = nd->intent.open.flags;
1352 if (nd->flags & LOOKUP_CREATE) {
1353 attr.ia_mode = nd->intent.open.create_mode;
1354 attr.ia_valid = ATTR_MODE;
1355 attr.ia_mode &= ~current_umask();
1357 open_flags &= ~(O_EXCL | O_CREAT);
1361 /* Open the file on the server */
1362 nfs_block_sillyrename(dentry->d_parent);
1363 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1364 if (IS_ERR(inode)) {
1365 nfs_unblock_sillyrename(dentry->d_parent);
1366 put_nfs_open_context(ctx);
1367 switch (PTR_ERR(inode)) {
1368 /* Make a negative dentry */
1370 d_add(dentry, NULL);
1373 /* This turned out not to be a regular file */
1377 if (!(nd->intent.open.flags & O_NOFOLLOW))
1382 res = ERR_CAST(inode);
1386 res = d_add_unique(dentry, inode);
1387 nfs_unblock_sillyrename(dentry->d_parent);
1389 dput(ctx->path.dentry);
1390 ctx->path.dentry = dget(res);
1393 err = nfs_intent_set_file(nd, ctx);
1397 return ERR_PTR(err);
1400 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1403 return nfs_lookup(dir, dentry, nd);
1406 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1408 struct dentry *parent = NULL;
1409 struct inode *inode;
1411 struct nfs_open_context *ctx;
1412 int openflags, ret = 0;
1414 if (nd->flags & LOOKUP_RCU)
1417 inode = dentry->d_inode;
1418 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1421 parent = dget_parent(dentry);
1422 dir = parent->d_inode;
1424 /* We can't create new files in nfs_open_revalidate(), so we
1425 * optimize away revalidation of negative dentries.
1427 if (inode == NULL) {
1428 if (!nfs_neg_need_reval(dir, dentry, nd))
1433 /* NFS only supports OPEN on regular files */
1434 if (!S_ISREG(inode->i_mode))
1436 openflags = nd->intent.open.flags;
1437 /* We cannot do exclusive creation on a positive dentry */
1438 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1440 /* We can't create new files, or truncate existing ones here */
1441 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1443 ctx = nameidata_to_nfs_open_context(dentry, nd);
1448 * Note: we're not holding inode->i_mutex and so may be racing with
1449 * operations that change the directory. We therefore save the
1450 * change attribute *before* we do the RPC call.
1452 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1453 if (IS_ERR(inode)) {
1454 ret = PTR_ERR(inode);
1467 if (inode != dentry->d_inode)
1470 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1471 ret = nfs_intent_set_file(nd, ctx);
1481 put_nfs_open_context(ctx);
1487 return nfs_lookup_revalidate(dentry, nd);
1490 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1491 struct nameidata *nd)
1493 struct nfs_open_context *ctx = NULL;
1498 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1499 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1501 attr.ia_mode = mode;
1502 attr.ia_valid = ATTR_MODE;
1504 if ((nd->flags & LOOKUP_CREATE) != 0) {
1505 open_flags = nd->intent.open.flags;
1507 ctx = nameidata_to_nfs_open_context(dentry, nd);
1508 error = PTR_ERR(ctx);
1513 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1517 error = nfs_intent_set_file(nd, ctx);
1524 put_nfs_open_context(ctx);
1531 #endif /* CONFIG_NFSV4 */
1534 * Code common to create, mkdir, and mknod.
1536 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1537 struct nfs_fattr *fattr)
1539 struct dentry *parent = dget_parent(dentry);
1540 struct inode *dir = parent->d_inode;
1541 struct inode *inode;
1542 int error = -EACCES;
1546 /* We may have been initialized further down */
1547 if (dentry->d_inode)
1549 if (fhandle->size == 0) {
1550 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1554 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1555 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1556 struct nfs_server *server = NFS_SB(dentry->d_sb);
1557 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1561 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1562 error = PTR_ERR(inode);
1565 d_add(dentry, inode);
1570 nfs_mark_for_revalidate(dir);
1576 * Following a failed create operation, we drop the dentry rather
1577 * than retain a negative dentry. This avoids a problem in the event
1578 * that the operation succeeded on the server, but an error in the
1579 * reply path made it appear to have failed.
1581 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1582 struct nameidata *nd)
1588 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1589 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1591 attr.ia_mode = mode;
1592 attr.ia_valid = ATTR_MODE;
1594 if ((nd->flags & LOOKUP_CREATE) != 0)
1595 open_flags = nd->intent.open.flags;
1597 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1607 * See comments for nfs_proc_create regarding failed operations.
1610 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1615 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1616 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1618 if (!new_valid_dev(rdev))
1621 attr.ia_mode = mode;
1622 attr.ia_valid = ATTR_MODE;
1624 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1634 * See comments for nfs_proc_create regarding failed operations.
1636 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1641 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1642 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1644 attr.ia_valid = ATTR_MODE;
1645 attr.ia_mode = mode | S_IFDIR;
1647 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1656 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1658 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1662 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1666 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1667 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1669 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1670 /* Ensure the VFS deletes this inode */
1671 if (error == 0 && dentry->d_inode != NULL)
1672 clear_nlink(dentry->d_inode);
1673 else if (error == -ENOENT)
1674 nfs_dentry_handle_enoent(dentry);
1680 * Remove a file after making sure there are no pending writes,
1681 * and after checking that the file has only one user.
1683 * We invalidate the attribute cache and free the inode prior to the operation
1684 * to avoid possible races if the server reuses the inode.
1686 static int nfs_safe_remove(struct dentry *dentry)
1688 struct inode *dir = dentry->d_parent->d_inode;
1689 struct inode *inode = dentry->d_inode;
1692 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1693 dentry->d_parent->d_name.name, dentry->d_name.name);
1695 /* If the dentry was sillyrenamed, we simply call d_delete() */
1696 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1701 if (inode != NULL) {
1702 nfs_inode_return_delegation(inode);
1703 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1704 /* The VFS may want to delete this inode */
1706 nfs_drop_nlink(inode);
1707 nfs_mark_for_revalidate(inode);
1709 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1710 if (error == -ENOENT)
1711 nfs_dentry_handle_enoent(dentry);
1716 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1717 * belongs to an active ".nfs..." file and we return -EBUSY.
1719 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1721 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1724 int need_rehash = 0;
1726 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1727 dir->i_ino, dentry->d_name.name);
1729 spin_lock(&dentry->d_lock);
1730 if (dentry->d_count > 1) {
1731 spin_unlock(&dentry->d_lock);
1732 /* Start asynchronous writeout of the inode */
1733 write_inode_now(dentry->d_inode, 0);
1734 error = nfs_sillyrename(dir, dentry);
1737 if (!d_unhashed(dentry)) {
1741 spin_unlock(&dentry->d_lock);
1742 error = nfs_safe_remove(dentry);
1743 if (!error || error == -ENOENT) {
1744 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1745 } else if (need_rehash)
1751 * To create a symbolic link, most file systems instantiate a new inode,
1752 * add a page to it containing the path, then write it out to the disk
1753 * using prepare_write/commit_write.
1755 * Unfortunately the NFS client can't create the in-core inode first
1756 * because it needs a file handle to create an in-core inode (see
1757 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1758 * symlink request has completed on the server.
1760 * So instead we allocate a raw page, copy the symname into it, then do
1761 * the SYMLINK request with the page as the buffer. If it succeeds, we
1762 * now have a new file handle and can instantiate an in-core NFS inode
1763 * and move the raw page into its mapping.
1765 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1767 struct pagevec lru_pvec;
1771 unsigned int pathlen = strlen(symname);
1774 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1775 dir->i_ino, dentry->d_name.name, symname);
1777 if (pathlen > PAGE_SIZE)
1778 return -ENAMETOOLONG;
1780 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1781 attr.ia_valid = ATTR_MODE;
1783 page = alloc_page(GFP_HIGHUSER);
1787 kaddr = kmap_atomic(page, KM_USER0);
1788 memcpy(kaddr, symname, pathlen);
1789 if (pathlen < PAGE_SIZE)
1790 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1791 kunmap_atomic(kaddr, KM_USER0);
1793 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1795 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1796 dir->i_sb->s_id, dir->i_ino,
1797 dentry->d_name.name, symname, error);
1804 * No big deal if we can't add this page to the page cache here.
1805 * READLINK will get the missing page from the server if needed.
1807 pagevec_init(&lru_pvec, 0);
1808 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1810 pagevec_add(&lru_pvec, page);
1811 pagevec_lru_add_file(&lru_pvec);
1812 SetPageUptodate(page);
1821 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1823 struct inode *inode = old_dentry->d_inode;
1826 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1827 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1828 dentry->d_parent->d_name.name, dentry->d_name.name);
1830 nfs_inode_return_delegation(inode);
1833 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1836 d_add(dentry, inode);
1843 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1844 * different file handle for the same inode after a rename (e.g. when
1845 * moving to a different directory). A fail-safe method to do so would
1846 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1847 * rename the old file using the sillyrename stuff. This way, the original
1848 * file in old_dir will go away when the last process iput()s the inode.
1852 * It actually works quite well. One needs to have the possibility for
1853 * at least one ".nfs..." file in each directory the file ever gets
1854 * moved or linked to which happens automagically with the new
1855 * implementation that only depends on the dcache stuff instead of
1856 * using the inode layer
1858 * Unfortunately, things are a little more complicated than indicated
1859 * above. For a cross-directory move, we want to make sure we can get
1860 * rid of the old inode after the operation. This means there must be
1861 * no pending writes (if it's a file), and the use count must be 1.
1862 * If these conditions are met, we can drop the dentries before doing
1865 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1866 struct inode *new_dir, struct dentry *new_dentry)
1868 struct inode *old_inode = old_dentry->d_inode;
1869 struct inode *new_inode = new_dentry->d_inode;
1870 struct dentry *dentry = NULL, *rehash = NULL;
1873 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1874 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1875 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1876 new_dentry->d_count);
1879 * For non-directories, check whether the target is busy and if so,
1880 * make a copy of the dentry and then do a silly-rename. If the
1881 * silly-rename succeeds, the copied dentry is hashed and becomes
1884 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1886 * To prevent any new references to the target during the
1887 * rename, we unhash the dentry in advance.
1889 if (!d_unhashed(new_dentry)) {
1891 rehash = new_dentry;
1894 if (new_dentry->d_count > 2) {
1897 /* copy the target dentry's name */
1898 dentry = d_alloc(new_dentry->d_parent,
1899 &new_dentry->d_name);
1903 /* silly-rename the existing target ... */
1904 err = nfs_sillyrename(new_dir, new_dentry);
1908 new_dentry = dentry;
1914 nfs_inode_return_delegation(old_inode);
1915 if (new_inode != NULL)
1916 nfs_inode_return_delegation(new_inode);
1918 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1919 new_dir, &new_dentry->d_name);
1920 nfs_mark_for_revalidate(old_inode);
1925 if (new_inode != NULL)
1926 nfs_drop_nlink(new_inode);
1927 d_move(old_dentry, new_dentry);
1928 nfs_set_verifier(new_dentry,
1929 nfs_save_change_attribute(new_dir));
1930 } else if (error == -ENOENT)
1931 nfs_dentry_handle_enoent(old_dentry);
1933 /* new dentry created? */
1939 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1940 static LIST_HEAD(nfs_access_lru_list);
1941 static atomic_long_t nfs_access_nr_entries;
1943 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1945 put_rpccred(entry->cred);
1947 smp_mb__before_atomic_dec();
1948 atomic_long_dec(&nfs_access_nr_entries);
1949 smp_mb__after_atomic_dec();
1952 static void nfs_access_free_list(struct list_head *head)
1954 struct nfs_access_entry *cache;
1956 while (!list_empty(head)) {
1957 cache = list_entry(head->next, struct nfs_access_entry, lru);
1958 list_del(&cache->lru);
1959 nfs_access_free_entry(cache);
1963 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1966 struct nfs_inode *nfsi, *next;
1967 struct nfs_access_entry *cache;
1969 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1970 return (nr_to_scan == 0) ? 0 : -1;
1972 spin_lock(&nfs_access_lru_lock);
1973 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1974 struct inode *inode;
1976 if (nr_to_scan-- == 0)
1978 inode = &nfsi->vfs_inode;
1979 spin_lock(&inode->i_lock);
1980 if (list_empty(&nfsi->access_cache_entry_lru))
1981 goto remove_lru_entry;
1982 cache = list_entry(nfsi->access_cache_entry_lru.next,
1983 struct nfs_access_entry, lru);
1984 list_move(&cache->lru, &head);
1985 rb_erase(&cache->rb_node, &nfsi->access_cache);
1986 if (!list_empty(&nfsi->access_cache_entry_lru))
1987 list_move_tail(&nfsi->access_cache_inode_lru,
1988 &nfs_access_lru_list);
1991 list_del_init(&nfsi->access_cache_inode_lru);
1992 smp_mb__before_clear_bit();
1993 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1994 smp_mb__after_clear_bit();
1996 spin_unlock(&inode->i_lock);
1998 spin_unlock(&nfs_access_lru_lock);
1999 nfs_access_free_list(&head);
2000 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2003 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2005 struct rb_root *root_node = &nfsi->access_cache;
2007 struct nfs_access_entry *entry;
2009 /* Unhook entries from the cache */
2010 while ((n = rb_first(root_node)) != NULL) {
2011 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2012 rb_erase(n, root_node);
2013 list_move(&entry->lru, head);
2015 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2018 void nfs_access_zap_cache(struct inode *inode)
2022 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2024 /* Remove from global LRU init */
2025 spin_lock(&nfs_access_lru_lock);
2026 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2027 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2029 spin_lock(&inode->i_lock);
2030 __nfs_access_zap_cache(NFS_I(inode), &head);
2031 spin_unlock(&inode->i_lock);
2032 spin_unlock(&nfs_access_lru_lock);
2033 nfs_access_free_list(&head);
2036 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2038 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2039 struct nfs_access_entry *entry;
2042 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2044 if (cred < entry->cred)
2046 else if (cred > entry->cred)
2054 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2056 struct nfs_inode *nfsi = NFS_I(inode);
2057 struct nfs_access_entry *cache;
2060 spin_lock(&inode->i_lock);
2061 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2063 cache = nfs_access_search_rbtree(inode, cred);
2066 if (!nfs_have_delegated_attributes(inode) &&
2067 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2069 res->jiffies = cache->jiffies;
2070 res->cred = cache->cred;
2071 res->mask = cache->mask;
2072 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2075 spin_unlock(&inode->i_lock);
2078 rb_erase(&cache->rb_node, &nfsi->access_cache);
2079 list_del(&cache->lru);
2080 spin_unlock(&inode->i_lock);
2081 nfs_access_free_entry(cache);
2084 spin_unlock(&inode->i_lock);
2085 nfs_access_zap_cache(inode);
2089 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2091 struct nfs_inode *nfsi = NFS_I(inode);
2092 struct rb_root *root_node = &nfsi->access_cache;
2093 struct rb_node **p = &root_node->rb_node;
2094 struct rb_node *parent = NULL;
2095 struct nfs_access_entry *entry;
2097 spin_lock(&inode->i_lock);
2098 while (*p != NULL) {
2100 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2102 if (set->cred < entry->cred)
2103 p = &parent->rb_left;
2104 else if (set->cred > entry->cred)
2105 p = &parent->rb_right;
2109 rb_link_node(&set->rb_node, parent, p);
2110 rb_insert_color(&set->rb_node, root_node);
2111 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2112 spin_unlock(&inode->i_lock);
2115 rb_replace_node(parent, &set->rb_node, root_node);
2116 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2117 list_del(&entry->lru);
2118 spin_unlock(&inode->i_lock);
2119 nfs_access_free_entry(entry);
2122 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2124 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2127 RB_CLEAR_NODE(&cache->rb_node);
2128 cache->jiffies = set->jiffies;
2129 cache->cred = get_rpccred(set->cred);
2130 cache->mask = set->mask;
2132 nfs_access_add_rbtree(inode, cache);
2134 /* Update accounting */
2135 smp_mb__before_atomic_inc();
2136 atomic_long_inc(&nfs_access_nr_entries);
2137 smp_mb__after_atomic_inc();
2139 /* Add inode to global LRU list */
2140 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2141 spin_lock(&nfs_access_lru_lock);
2142 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2143 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2144 &nfs_access_lru_list);
2145 spin_unlock(&nfs_access_lru_lock);
2149 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2151 struct nfs_access_entry cache;
2154 status = nfs_access_get_cached(inode, cred, &cache);
2158 /* Be clever: ask server to check for all possible rights */
2159 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2161 cache.jiffies = jiffies;
2162 status = NFS_PROTO(inode)->access(inode, &cache);
2164 if (status == -ESTALE) {
2165 nfs_zap_caches(inode);
2166 if (!S_ISDIR(inode->i_mode))
2167 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2171 nfs_access_add_cache(inode, &cache);
2173 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2178 static int nfs_open_permission_mask(int openflags)
2182 if (openflags & FMODE_READ)
2184 if (openflags & FMODE_WRITE)
2186 if (openflags & FMODE_EXEC)
2191 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2193 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2196 int nfs_permission(struct inode *inode, int mask, unsigned int flags)
2198 struct rpc_cred *cred;
2201 if (flags & IPERM_FLAG_RCU)
2204 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2206 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2208 /* Is this sys_access() ? */
2209 if (mask & (MAY_ACCESS | MAY_CHDIR))
2212 switch (inode->i_mode & S_IFMT) {
2216 /* NFSv4 has atomic_open... */
2217 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2218 && (mask & MAY_OPEN)
2219 && !(mask & MAY_EXEC))
2224 * Optimize away all write operations, since the server
2225 * will check permissions when we perform the op.
2227 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2232 if (!NFS_PROTO(inode)->access)
2235 cred = rpc_lookup_cred();
2236 if (!IS_ERR(cred)) {
2237 res = nfs_do_access(inode, cred, mask);
2240 res = PTR_ERR(cred);
2242 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2245 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2246 inode->i_sb->s_id, inode->i_ino, mask, res);
2249 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2251 res = generic_permission(inode, mask, flags, NULL);
2257 * version-control: t
2258 * kept-new-versions: 5