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/vmalloc.h>
38 #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 *, int);
58 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 static int nfs_readdir_clear_array(struct page*, gfp_t);
61 const struct file_operations nfs_dir_operations = {
62 .llseek = nfs_llseek_dir,
63 .read = generic_read_dir,
64 .readdir = nfs_readdir,
66 .release = nfs_release,
67 .fsync = nfs_fsync_dir,
70 const struct inode_operations nfs_dir_inode_operations = {
75 .symlink = nfs_symlink,
80 .permission = nfs_permission,
81 .getattr = nfs_getattr,
82 .setattr = nfs_setattr,
85 const struct address_space_operations nfs_dir_addr_space_ops = {
86 .releasepage = nfs_readdir_clear_array,
90 const struct inode_operations nfs3_dir_inode_operations = {
95 .symlink = nfs_symlink,
100 .permission = nfs_permission,
101 .getattr = nfs_getattr,
102 .setattr = nfs_setattr,
103 .listxattr = nfs3_listxattr,
104 .getxattr = nfs3_getxattr,
105 .setxattr = nfs3_setxattr,
106 .removexattr = nfs3_removexattr,
108 #endif /* CONFIG_NFS_V3 */
112 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
113 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
114 const struct inode_operations nfs4_dir_inode_operations = {
115 .create = nfs_open_create,
116 .lookup = nfs_atomic_lookup,
118 .unlink = nfs_unlink,
119 .symlink = nfs_symlink,
123 .rename = nfs_rename,
124 .permission = nfs_permission,
125 .getattr = nfs_getattr,
126 .setattr = nfs_setattr,
127 .getxattr = nfs4_getxattr,
128 .setxattr = nfs4_setxattr,
129 .listxattr = nfs4_listxattr,
132 #endif /* CONFIG_NFS_V4 */
138 nfs_opendir(struct inode *inode, struct file *filp)
142 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
143 filp->f_path.dentry->d_parent->d_name.name,
144 filp->f_path.dentry->d_name.name);
146 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
148 /* Call generic open code in order to cache credentials */
149 res = nfs_open(inode, filp);
150 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
151 /* This is a mountpoint, so d_revalidate will never
152 * have been called, so we need to refresh the
153 * inode (for close-open consistency) ourselves.
155 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
160 struct nfs_cache_array_entry {
166 struct nfs_cache_array {
170 struct nfs_cache_array_entry array[0];
173 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
175 typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int);
179 unsigned long page_index;
181 loff_t current_index;
182 decode_dirent_t decode;
184 unsigned long timestamp;
185 unsigned long gencount;
186 unsigned int cache_entry_index;
189 } nfs_readdir_descriptor_t;
192 * The caller is responsible for calling nfs_readdir_release_array(page)
195 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
199 return ERR_PTR(-EIO);
202 return ERR_PTR(-ENOMEM);
207 void nfs_readdir_release_array(struct page *page)
213 * we are freeing strings created by nfs_add_to_readdir_array()
216 int nfs_readdir_clear_array(struct page *page, gfp_t mask)
218 struct nfs_cache_array *array = nfs_readdir_get_array(page);
222 return PTR_ERR(array);
223 for (i = 0; i < array->size; i++)
224 kfree(array->array[i].string.name);
225 nfs_readdir_release_array(page);
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)
241 string->hash = full_name_hash(name, len);
246 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
248 struct nfs_cache_array *array = nfs_readdir_get_array(page);
249 struct nfs_cache_array_entry *cache_entry;
253 return PTR_ERR(array);
255 if (array->size >= MAX_READDIR_ARRAY)
258 cache_entry = &array->array[array->size];
259 cache_entry->cookie = entry->prev_cookie;
260 cache_entry->ino = entry->ino;
261 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
264 array->last_cookie = entry->cookie;
267 array->eof_index = array->size;
269 nfs_readdir_release_array(page);
274 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
276 loff_t diff = desc->file->f_pos - desc->current_index;
281 if (diff >= array->size) {
282 if (array->eof_index >= 0)
284 desc->current_index += array->size;
288 index = (unsigned int)diff;
289 *desc->dir_cookie = array->array[index].cookie;
290 desc->cache_entry_index = index;
298 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
301 int status = -EAGAIN;
303 for (i = 0; i < array->size; i++) {
304 if (array->array[i].cookie == *desc->dir_cookie) {
305 desc->cache_entry_index = i;
310 if (i == array->eof_index) {
312 status = -EBADCOOKIE;
319 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
321 struct nfs_cache_array *array;
322 int status = -EBADCOOKIE;
324 if (desc->dir_cookie == NULL)
327 array = nfs_readdir_get_array(desc->page);
329 status = PTR_ERR(array);
333 if (*desc->dir_cookie == 0)
334 status = nfs_readdir_search_for_pos(array, desc);
336 status = nfs_readdir_search_for_cookie(array, desc);
338 nfs_readdir_release_array(desc->page);
343 /* Fill a page with xdr information before transferring to the cache page */
345 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
346 struct nfs_entry *entry, struct file *file, struct inode *inode)
348 struct rpc_cred *cred = nfs_file_cred(file);
349 unsigned long timestamp, gencount;
354 gencount = nfs_inc_attr_generation_counter();
355 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
356 NFS_SERVER(inode)->dtsize, desc->plus);
358 /* We requested READDIRPLUS, but the server doesn't grok it */
359 if (error == -ENOTSUPP && desc->plus) {
360 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
361 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
367 desc->timestamp = timestamp;
368 desc->gencount = gencount;
373 /* Fill in an entry based on the xdr code stored in desc->page */
375 int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream)
377 __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus);
381 entry->fattr->time_start = desc->timestamp;
382 entry->fattr->gencount = desc->gencount;
387 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
389 struct nfs_inode *node;
390 if (dentry->d_inode == NULL)
392 node = NFS_I(dentry->d_inode);
393 if (node->fh.size != entry->fh->size)
395 if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 0)
403 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
405 struct qstr filename = {
409 struct dentry *dentry;
410 struct dentry *alias;
411 struct inode *dir = parent->d_inode;
414 if (filename.name[0] == '.') {
415 if (filename.len == 1)
417 if (filename.len == 2 && filename.name[1] == '.')
420 filename.hash = full_name_hash(filename.name, filename.len);
422 dentry = d_lookup(parent, &filename);
423 if (dentry != NULL) {
424 if (nfs_same_file(dentry, entry)) {
425 nfs_refresh_inode(dentry->d_inode, entry->fattr);
433 dentry = d_alloc(parent, &filename);
437 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
438 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
442 alias = d_materialise_unique(dentry, inode);
446 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
449 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
455 /* Perform conversion from xdr to cache array */
457 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
458 void *xdr_page, struct page *page, unsigned int buflen)
460 struct xdr_stream stream;
462 __be32 *ptr = xdr_page;
464 struct nfs_cache_array *array;
466 buf.head->iov_base = xdr_page;
467 buf.head->iov_len = buflen;
468 buf.tail->iov_len = 0;
471 buf.buflen = buf.head->iov_len;
472 buf.len = buf.head->iov_len;
474 xdr_init_decode(&stream, &buf, ptr);
478 status = xdr_decode(desc, entry, &stream);
480 if (status == -EAGAIN)
486 nfs_prime_dcache(desc->file->f_path.dentry, entry);
488 status = nfs_readdir_add_to_array(entry, page);
491 } while (!entry->eof);
493 if (status == -EBADCOOKIE && entry->eof) {
494 array = nfs_readdir_get_array(page);
495 if (!IS_ERR(array)) {
496 array->eof_index = array->size;
498 nfs_readdir_release_array(page);
505 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
508 for (i = 0; i < npages; i++)
513 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
516 vm_unmap_ram(ptr, npages);
517 nfs_readdir_free_pagearray(pages, npages);
521 * nfs_readdir_large_page will allocate pages that must be freed with a call
522 * to nfs_readdir_free_large_page
525 void *nfs_readdir_large_page(struct page **pages, unsigned int npages)
530 for (i = 0; i < npages; i++) {
531 struct page *page = alloc_page(GFP_KERNEL);
537 ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL);
538 if (!IS_ERR_OR_NULL(ptr))
541 nfs_readdir_free_pagearray(pages, i);
546 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
548 struct page *pages[NFS_MAX_READDIR_PAGES];
549 void *pages_ptr = NULL;
550 struct nfs_entry entry;
551 struct file *file = desc->file;
552 struct nfs_cache_array *array;
553 int status = -ENOMEM;
554 unsigned int array_size = ARRAY_SIZE(pages);
556 entry.prev_cookie = 0;
557 entry.cookie = *desc->dir_cookie;
559 entry.fh = nfs_alloc_fhandle();
560 entry.fattr = nfs_alloc_fattr();
561 if (entry.fh == NULL || entry.fattr == NULL)
564 array = nfs_readdir_get_array(page);
566 status = PTR_ERR(array);
569 memset(array, 0, sizeof(struct nfs_cache_array));
570 array->eof_index = -1;
572 pages_ptr = nfs_readdir_large_page(pages, array_size);
574 goto out_release_array;
577 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
582 status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen);
584 if (status == -ENOSPC)
588 } while (array->eof_index < 0);
590 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
592 nfs_readdir_release_array(page);
594 nfs_free_fattr(entry.fattr);
595 nfs_free_fhandle(entry.fh);
600 * Now we cache directories properly, by converting xdr information
601 * to an array that can be used for lookups later. This results in
602 * fewer cache pages, since we can store more information on each page.
603 * We only need to convert from xdr once so future lookups are much simpler
606 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
608 struct inode *inode = desc->file->f_path.dentry->d_inode;
611 ret = nfs_readdir_xdr_to_array(desc, page, inode);
614 SetPageUptodate(page);
616 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
617 /* Should never happen */
618 nfs_zap_mapping(inode, inode->i_mapping);
628 void cache_page_release(nfs_readdir_descriptor_t *desc)
630 page_cache_release(desc->page);
635 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
637 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
638 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
642 * Returns 0 if desc->dir_cookie was found on page desc->page_index
645 int find_cache_page(nfs_readdir_descriptor_t *desc)
649 desc->page = get_cache_page(desc);
650 if (IS_ERR(desc->page))
651 return PTR_ERR(desc->page);
653 res = nfs_readdir_search_array(desc);
656 cache_page_release(desc);
660 /* Search for desc->dir_cookie from the beginning of the page cache */
662 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
666 if (desc->page_index == 0)
667 desc->current_index = 0;
669 res = find_cache_page(desc);
677 static inline unsigned int dt_type(struct inode *inode)
679 return (inode->i_mode >> 12) & 15;
683 * Once we've found the start of the dirent within a page: fill 'er up...
686 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
689 struct file *file = desc->file;
692 struct nfs_cache_array *array = NULL;
693 unsigned int d_type = DT_UNKNOWN;
694 struct dentry *dentry = NULL;
696 array = nfs_readdir_get_array(desc->page);
698 return PTR_ERR(array);
700 for (i = desc->cache_entry_index; i < array->size; i++) {
703 res = filldir(dirent, array->array[i].string.name,
704 array->array[i].string.len, file->f_pos,
705 nfs_compat_user_ino64(array->array[i].ino), d_type);
709 desc->cache_entry_index = i;
710 if (i < (array->size-1))
711 *desc->dir_cookie = array->array[i+1].cookie;
713 *desc->dir_cookie = array->last_cookie;
715 if (i == array->eof_index)
718 nfs_readdir_release_array(desc->page);
719 cache_page_release(desc);
722 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
723 (unsigned long long)*desc->dir_cookie, res);
728 * If we cannot find a cookie in our cache, we suspect that this is
729 * because it points to a deleted file, so we ask the server to return
730 * whatever it thinks is the next entry. We then feed this to filldir.
731 * If all goes well, we should then be able to find our way round the
732 * cache on the next call to readdir_search_pagecache();
734 * NOTE: we cannot add the anonymous page to the pagecache because
735 * the data it contains might not be page aligned. Besides,
736 * we should already have a complete representation of the
737 * directory in the page cache by the time we get here.
740 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
743 struct page *page = NULL;
745 struct inode *inode = desc->file->f_path.dentry->d_inode;
747 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
748 (unsigned long long)*desc->dir_cookie);
750 page = alloc_page(GFP_HIGHUSER);
756 if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) {
761 desc->page_index = 0;
763 status = nfs_do_filldir(desc, dirent, filldir);
766 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
770 cache_page_release(desc);
774 /* The file offset position represents the dirent entry number. A
775 last cookie cache takes care of the common case of reading the
778 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
780 struct dentry *dentry = filp->f_path.dentry;
781 struct inode *inode = dentry->d_inode;
782 nfs_readdir_descriptor_t my_desc,
786 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
787 dentry->d_parent->d_name.name, dentry->d_name.name,
788 (long long)filp->f_pos);
789 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
792 * filp->f_pos points to the dirent entry number.
793 * *desc->dir_cookie has the cookie for the next entry. We have
794 * to either find the entry with the appropriate number or
795 * revalidate the cookie.
797 memset(desc, 0, sizeof(*desc));
800 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
801 desc->decode = NFS_PROTO(inode)->decode_dirent;
802 desc->plus = NFS_USE_READDIRPLUS(inode);
804 nfs_block_sillyrename(dentry);
805 res = nfs_revalidate_mapping(inode, filp->f_mapping);
809 while (desc->eof != 1) {
810 res = readdir_search_pagecache(desc);
812 if (res == -EBADCOOKIE) {
813 /* This means either end of directory */
814 if (*desc->dir_cookie && desc->eof == 0) {
815 /* Or that the server has 'lost' a cookie */
816 res = uncached_readdir(desc, dirent, filldir);
823 if (res == -ETOOSMALL && desc->plus) {
824 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
825 nfs_zap_caches(inode);
826 desc->page_index = 0;
834 res = nfs_do_filldir(desc, dirent, filldir);
841 nfs_unblock_sillyrename(dentry);
844 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
845 dentry->d_parent->d_name.name, dentry->d_name.name,
850 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
852 struct dentry *dentry = filp->f_path.dentry;
853 struct inode *inode = dentry->d_inode;
855 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
856 dentry->d_parent->d_name.name,
860 mutex_lock(&inode->i_mutex);
863 offset += filp->f_pos;
871 if (offset != filp->f_pos) {
872 filp->f_pos = offset;
873 nfs_file_open_context(filp)->dir_cookie = 0;
876 mutex_unlock(&inode->i_mutex);
881 * All directory operations under NFS are synchronous, so fsync()
882 * is a dummy operation.
884 static int nfs_fsync_dir(struct file *filp, int datasync)
886 struct dentry *dentry = filp->f_path.dentry;
888 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
889 dentry->d_parent->d_name.name, dentry->d_name.name,
892 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
897 * nfs_force_lookup_revalidate - Mark the directory as having changed
898 * @dir - pointer to directory inode
900 * This forces the revalidation code in nfs_lookup_revalidate() to do a
901 * full lookup on all child dentries of 'dir' whenever a change occurs
902 * on the server that might have invalidated our dcache.
904 * The caller should be holding dir->i_lock
906 void nfs_force_lookup_revalidate(struct inode *dir)
908 NFS_I(dir)->cache_change_attribute++;
912 * A check for whether or not the parent directory has changed.
913 * In the case it has, we assume that the dentries are untrustworthy
914 * and may need to be looked up again.
916 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
920 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
922 if (!nfs_verify_change_attribute(dir, dentry->d_time))
924 /* Revalidate nfsi->cache_change_attribute before we declare a match */
925 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
927 if (!nfs_verify_change_attribute(dir, dentry->d_time))
933 * Return the intent data that applies to this particular path component
935 * Note that the current set of intents only apply to the very last
936 * component of the path.
937 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
939 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
941 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
943 return nd->flags & mask;
947 * Use intent information to check whether or not we're going to do
948 * an O_EXCL create using this path component.
950 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
952 if (NFS_PROTO(dir)->version == 2)
954 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
958 * Inode and filehandle revalidation for lookups.
960 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
961 * or if the intent information indicates that we're about to open this
962 * particular file and the "nocto" mount flag is not set.
966 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
968 struct nfs_server *server = NFS_SERVER(inode);
970 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
973 /* VFS wants an on-the-wire revalidation */
974 if (nd->flags & LOOKUP_REVAL)
976 /* This is an open(2) */
977 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
978 !(server->flags & NFS_MOUNT_NOCTO) &&
979 (S_ISREG(inode->i_mode) ||
980 S_ISDIR(inode->i_mode)))
984 return nfs_revalidate_inode(server, inode);
986 return __nfs_revalidate_inode(server, inode);
990 * We judge how long we want to trust negative
991 * dentries by looking at the parent inode mtime.
993 * If parent mtime has changed, we revalidate, else we wait for a
994 * period corresponding to the parent's attribute cache timeout value.
997 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
998 struct nameidata *nd)
1000 /* Don't revalidate a negative dentry if we're creating a new file */
1001 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1003 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1005 return !nfs_check_verifier(dir, dentry);
1009 * This is called every time the dcache has a lookup hit,
1010 * and we should check whether we can really trust that
1013 * NOTE! The hit can be a negative hit too, don't assume
1016 * If the parent directory is seen to have changed, we throw out the
1017 * cached dentry and do a new lookup.
1019 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
1022 struct inode *inode;
1023 struct dentry *parent;
1024 struct nfs_fh *fhandle = NULL;
1025 struct nfs_fattr *fattr = NULL;
1028 parent = dget_parent(dentry);
1029 dir = parent->d_inode;
1030 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1031 inode = dentry->d_inode;
1034 if (nfs_neg_need_reval(dir, dentry, nd))
1039 if (is_bad_inode(inode)) {
1040 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1041 __func__, dentry->d_parent->d_name.name,
1042 dentry->d_name.name);
1046 if (nfs_have_delegation(inode, FMODE_READ))
1047 goto out_set_verifier;
1049 /* Force a full look up iff the parent directory has changed */
1050 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1051 if (nfs_lookup_verify_inode(inode, nd))
1052 goto out_zap_parent;
1056 if (NFS_STALE(inode))
1060 fhandle = nfs_alloc_fhandle();
1061 fattr = nfs_alloc_fattr();
1062 if (fhandle == NULL || fattr == NULL)
1065 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1068 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1070 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1073 nfs_free_fattr(fattr);
1074 nfs_free_fhandle(fhandle);
1076 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1079 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1080 __func__, dentry->d_parent->d_name.name,
1081 dentry->d_name.name);
1084 nfs_zap_caches(dir);
1086 nfs_mark_for_revalidate(dir);
1087 if (inode && S_ISDIR(inode->i_mode)) {
1088 /* Purge readdir caches. */
1089 nfs_zap_caches(inode);
1090 /* If we have submounts, don't unhash ! */
1091 if (have_submounts(dentry))
1093 if (dentry->d_flags & DCACHE_DISCONNECTED)
1095 shrink_dcache_parent(dentry);
1098 nfs_free_fattr(fattr);
1099 nfs_free_fhandle(fhandle);
1101 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1102 __func__, dentry->d_parent->d_name.name,
1103 dentry->d_name.name);
1106 nfs_free_fattr(fattr);
1107 nfs_free_fhandle(fhandle);
1109 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1110 __func__, dentry->d_parent->d_name.name,
1111 dentry->d_name.name, error);
1116 * This is called from dput() when d_count is going to 0.
1118 static int nfs_dentry_delete(struct dentry *dentry)
1120 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1121 dentry->d_parent->d_name.name, dentry->d_name.name,
1124 /* Unhash any dentry with a stale inode */
1125 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1128 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1129 /* Unhash it, so that ->d_iput() would be called */
1132 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1133 /* Unhash it, so that ancestors of killed async unlink
1134 * files will be cleaned up during umount */
1141 static void nfs_drop_nlink(struct inode *inode)
1143 spin_lock(&inode->i_lock);
1144 if (inode->i_nlink > 0)
1146 spin_unlock(&inode->i_lock);
1150 * Called when the dentry loses inode.
1151 * We use it to clean up silly-renamed files.
1153 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1155 if (S_ISDIR(inode->i_mode))
1156 /* drop any readdir cache as it could easily be old */
1157 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1159 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1161 nfs_complete_unlink(dentry, inode);
1166 const struct dentry_operations nfs_dentry_operations = {
1167 .d_revalidate = nfs_lookup_revalidate,
1168 .d_delete = nfs_dentry_delete,
1169 .d_iput = nfs_dentry_iput,
1172 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1175 struct dentry *parent;
1176 struct inode *inode = NULL;
1177 struct nfs_fh *fhandle = NULL;
1178 struct nfs_fattr *fattr = NULL;
1181 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1182 dentry->d_parent->d_name.name, dentry->d_name.name);
1183 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1185 res = ERR_PTR(-ENAMETOOLONG);
1186 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1189 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1192 * If we're doing an exclusive create, optimize away the lookup
1193 * but don't hash the dentry.
1195 if (nfs_is_exclusive_create(dir, nd)) {
1196 d_instantiate(dentry, NULL);
1201 res = ERR_PTR(-ENOMEM);
1202 fhandle = nfs_alloc_fhandle();
1203 fattr = nfs_alloc_fattr();
1204 if (fhandle == NULL || fattr == NULL)
1207 parent = dentry->d_parent;
1208 /* Protect against concurrent sillydeletes */
1209 nfs_block_sillyrename(parent);
1210 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1211 if (error == -ENOENT)
1214 res = ERR_PTR(error);
1215 goto out_unblock_sillyrename;
1217 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1218 res = (struct dentry *)inode;
1220 goto out_unblock_sillyrename;
1223 res = d_materialise_unique(dentry, inode);
1226 goto out_unblock_sillyrename;
1229 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1230 out_unblock_sillyrename:
1231 nfs_unblock_sillyrename(parent);
1233 nfs_free_fattr(fattr);
1234 nfs_free_fhandle(fhandle);
1238 #ifdef CONFIG_NFS_V4
1239 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1241 const struct dentry_operations nfs4_dentry_operations = {
1242 .d_revalidate = nfs_open_revalidate,
1243 .d_delete = nfs_dentry_delete,
1244 .d_iput = nfs_dentry_iput,
1248 * Use intent information to determine whether we need to substitute
1249 * the NFSv4-style stateful OPEN for the LOOKUP call
1251 static int is_atomic_open(struct nameidata *nd)
1253 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1255 /* NFS does not (yet) have a stateful open for directories */
1256 if (nd->flags & LOOKUP_DIRECTORY)
1258 /* Are we trying to write to a read only partition? */
1259 if (__mnt_is_readonly(nd->path.mnt) &&
1260 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1265 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1267 struct path path = {
1268 .mnt = nd->path.mnt,
1271 struct nfs_open_context *ctx;
1272 struct rpc_cred *cred;
1273 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1275 cred = rpc_lookup_cred();
1277 return ERR_CAST(cred);
1278 ctx = alloc_nfs_open_context(&path, cred, fmode);
1281 return ERR_PTR(-ENOMEM);
1285 static int do_open(struct inode *inode, struct file *filp)
1287 nfs_fscache_set_inode_cookie(inode, filp);
1291 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1296 /* If the open_intent is for execute, we have an extra check to make */
1297 if (ctx->mode & FMODE_EXEC) {
1298 ret = nfs_may_open(ctx->path.dentry->d_inode,
1300 nd->intent.open.flags);
1304 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1306 ret = PTR_ERR(filp);
1308 nfs_file_set_open_context(filp, ctx);
1310 put_nfs_open_context(ctx);
1314 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1316 struct nfs_open_context *ctx;
1318 struct dentry *res = NULL;
1319 struct inode *inode;
1323 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1324 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1326 /* Check that we are indeed trying to open this file */
1327 if (!is_atomic_open(nd))
1330 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1331 res = ERR_PTR(-ENAMETOOLONG);
1334 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1336 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1338 if (nd->flags & LOOKUP_EXCL) {
1339 d_instantiate(dentry, NULL);
1343 ctx = nameidata_to_nfs_open_context(dentry, nd);
1344 res = ERR_CAST(ctx);
1348 open_flags = nd->intent.open.flags;
1349 if (nd->flags & LOOKUP_CREATE) {
1350 attr.ia_mode = nd->intent.open.create_mode;
1351 attr.ia_valid = ATTR_MODE;
1352 if (!IS_POSIXACL(dir))
1353 attr.ia_mode &= ~current_umask();
1355 open_flags &= ~(O_EXCL | O_CREAT);
1359 /* Open the file on the server */
1360 nfs_block_sillyrename(dentry->d_parent);
1361 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1362 if (IS_ERR(inode)) {
1363 nfs_unblock_sillyrename(dentry->d_parent);
1364 put_nfs_open_context(ctx);
1365 switch (PTR_ERR(inode)) {
1366 /* Make a negative dentry */
1368 d_add(dentry, NULL);
1371 /* This turned out not to be a regular file */
1375 if (!(nd->intent.open.flags & O_NOFOLLOW))
1380 res = ERR_CAST(inode);
1384 res = d_add_unique(dentry, inode);
1385 nfs_unblock_sillyrename(dentry->d_parent);
1387 dput(ctx->path.dentry);
1388 ctx->path.dentry = dget(res);
1391 err = nfs_intent_set_file(nd, ctx);
1395 return ERR_PTR(err);
1398 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1401 return nfs_lookup(dir, dentry, nd);
1404 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1406 struct dentry *parent = NULL;
1407 struct inode *inode = dentry->d_inode;
1409 struct nfs_open_context *ctx;
1410 int openflags, ret = 0;
1412 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1415 parent = dget_parent(dentry);
1416 dir = parent->d_inode;
1418 /* We can't create new files in nfs_open_revalidate(), so we
1419 * optimize away revalidation of negative dentries.
1421 if (inode == NULL) {
1422 if (!nfs_neg_need_reval(dir, dentry, nd))
1427 /* NFS only supports OPEN on regular files */
1428 if (!S_ISREG(inode->i_mode))
1430 openflags = nd->intent.open.flags;
1431 /* We cannot do exclusive creation on a positive dentry */
1432 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1434 /* We can't create new files, or truncate existing ones here */
1435 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1437 ctx = nameidata_to_nfs_open_context(dentry, nd);
1442 * Note: we're not holding inode->i_mutex and so may be racing with
1443 * operations that change the directory. We therefore save the
1444 * change attribute *before* we do the RPC call.
1446 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1447 if (IS_ERR(inode)) {
1448 ret = PTR_ERR(inode);
1461 if (inode != dentry->d_inode)
1464 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1465 ret = nfs_intent_set_file(nd, ctx);
1475 put_nfs_open_context(ctx);
1481 return nfs_lookup_revalidate(dentry, nd);
1484 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1485 struct nameidata *nd)
1487 struct nfs_open_context *ctx = NULL;
1492 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1493 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1495 attr.ia_mode = mode;
1496 attr.ia_valid = ATTR_MODE;
1498 if ((nd->flags & LOOKUP_CREATE) != 0) {
1499 open_flags = nd->intent.open.flags;
1501 ctx = nameidata_to_nfs_open_context(dentry, nd);
1502 error = PTR_ERR(ctx);
1507 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1511 error = nfs_intent_set_file(nd, ctx);
1518 put_nfs_open_context(ctx);
1525 #endif /* CONFIG_NFSV4 */
1528 * Code common to create, mkdir, and mknod.
1530 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1531 struct nfs_fattr *fattr)
1533 struct dentry *parent = dget_parent(dentry);
1534 struct inode *dir = parent->d_inode;
1535 struct inode *inode;
1536 int error = -EACCES;
1540 /* We may have been initialized further down */
1541 if (dentry->d_inode)
1543 if (fhandle->size == 0) {
1544 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1548 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1549 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1550 struct nfs_server *server = NFS_SB(dentry->d_sb);
1551 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1555 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1556 error = PTR_ERR(inode);
1559 d_add(dentry, inode);
1564 nfs_mark_for_revalidate(dir);
1570 * Following a failed create operation, we drop the dentry rather
1571 * than retain a negative dentry. This avoids a problem in the event
1572 * that the operation succeeded on the server, but an error in the
1573 * reply path made it appear to have failed.
1575 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1576 struct nameidata *nd)
1581 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1582 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1584 attr.ia_mode = mode;
1585 attr.ia_valid = ATTR_MODE;
1587 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1597 * See comments for nfs_proc_create regarding failed operations.
1600 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1605 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1606 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1608 if (!new_valid_dev(rdev))
1611 attr.ia_mode = mode;
1612 attr.ia_valid = ATTR_MODE;
1614 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1624 * See comments for nfs_proc_create regarding failed operations.
1626 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1631 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1632 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1634 attr.ia_valid = ATTR_MODE;
1635 attr.ia_mode = mode | S_IFDIR;
1637 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1646 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1648 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1652 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1656 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1657 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1659 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1660 /* Ensure the VFS deletes this inode */
1661 if (error == 0 && dentry->d_inode != NULL)
1662 clear_nlink(dentry->d_inode);
1663 else if (error == -ENOENT)
1664 nfs_dentry_handle_enoent(dentry);
1670 * Remove a file after making sure there are no pending writes,
1671 * and after checking that the file has only one user.
1673 * We invalidate the attribute cache and free the inode prior to the operation
1674 * to avoid possible races if the server reuses the inode.
1676 static int nfs_safe_remove(struct dentry *dentry)
1678 struct inode *dir = dentry->d_parent->d_inode;
1679 struct inode *inode = dentry->d_inode;
1682 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1683 dentry->d_parent->d_name.name, dentry->d_name.name);
1685 /* If the dentry was sillyrenamed, we simply call d_delete() */
1686 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1691 if (inode != NULL) {
1692 nfs_inode_return_delegation(inode);
1693 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1694 /* The VFS may want to delete this inode */
1696 nfs_drop_nlink(inode);
1697 nfs_mark_for_revalidate(inode);
1699 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1700 if (error == -ENOENT)
1701 nfs_dentry_handle_enoent(dentry);
1706 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1707 * belongs to an active ".nfs..." file and we return -EBUSY.
1709 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1711 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1714 int need_rehash = 0;
1716 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1717 dir->i_ino, dentry->d_name.name);
1719 spin_lock(&dcache_lock);
1720 spin_lock(&dentry->d_lock);
1721 if (atomic_read(&dentry->d_count) > 1) {
1722 spin_unlock(&dentry->d_lock);
1723 spin_unlock(&dcache_lock);
1724 /* Start asynchronous writeout of the inode */
1725 write_inode_now(dentry->d_inode, 0);
1726 error = nfs_sillyrename(dir, dentry);
1729 if (!d_unhashed(dentry)) {
1733 spin_unlock(&dentry->d_lock);
1734 spin_unlock(&dcache_lock);
1735 error = nfs_safe_remove(dentry);
1736 if (!error || error == -ENOENT) {
1737 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1738 } else if (need_rehash)
1744 * To create a symbolic link, most file systems instantiate a new inode,
1745 * add a page to it containing the path, then write it out to the disk
1746 * using prepare_write/commit_write.
1748 * Unfortunately the NFS client can't create the in-core inode first
1749 * because it needs a file handle to create an in-core inode (see
1750 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1751 * symlink request has completed on the server.
1753 * So instead we allocate a raw page, copy the symname into it, then do
1754 * the SYMLINK request with the page as the buffer. If it succeeds, we
1755 * now have a new file handle and can instantiate an in-core NFS inode
1756 * and move the raw page into its mapping.
1758 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1760 struct pagevec lru_pvec;
1764 unsigned int pathlen = strlen(symname);
1767 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1768 dir->i_ino, dentry->d_name.name, symname);
1770 if (pathlen > PAGE_SIZE)
1771 return -ENAMETOOLONG;
1773 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1774 attr.ia_valid = ATTR_MODE;
1776 page = alloc_page(GFP_HIGHUSER);
1780 kaddr = kmap_atomic(page, KM_USER0);
1781 memcpy(kaddr, symname, pathlen);
1782 if (pathlen < PAGE_SIZE)
1783 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1784 kunmap_atomic(kaddr, KM_USER0);
1786 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1788 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1789 dir->i_sb->s_id, dir->i_ino,
1790 dentry->d_name.name, symname, error);
1797 * No big deal if we can't add this page to the page cache here.
1798 * READLINK will get the missing page from the server if needed.
1800 pagevec_init(&lru_pvec, 0);
1801 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1803 pagevec_add(&lru_pvec, page);
1804 pagevec_lru_add_file(&lru_pvec);
1805 SetPageUptodate(page);
1814 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1816 struct inode *inode = old_dentry->d_inode;
1819 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1820 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1821 dentry->d_parent->d_name.name, dentry->d_name.name);
1823 nfs_inode_return_delegation(inode);
1826 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1829 d_add(dentry, inode);
1836 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1837 * different file handle for the same inode after a rename (e.g. when
1838 * moving to a different directory). A fail-safe method to do so would
1839 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1840 * rename the old file using the sillyrename stuff. This way, the original
1841 * file in old_dir will go away when the last process iput()s the inode.
1845 * It actually works quite well. One needs to have the possibility for
1846 * at least one ".nfs..." file in each directory the file ever gets
1847 * moved or linked to which happens automagically with the new
1848 * implementation that only depends on the dcache stuff instead of
1849 * using the inode layer
1851 * Unfortunately, things are a little more complicated than indicated
1852 * above. For a cross-directory move, we want to make sure we can get
1853 * rid of the old inode after the operation. This means there must be
1854 * no pending writes (if it's a file), and the use count must be 1.
1855 * If these conditions are met, we can drop the dentries before doing
1858 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1859 struct inode *new_dir, struct dentry *new_dentry)
1861 struct inode *old_inode = old_dentry->d_inode;
1862 struct inode *new_inode = new_dentry->d_inode;
1863 struct dentry *dentry = NULL, *rehash = NULL;
1866 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1867 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1868 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1869 atomic_read(&new_dentry->d_count));
1872 * For non-directories, check whether the target is busy and if so,
1873 * make a copy of the dentry and then do a silly-rename. If the
1874 * silly-rename succeeds, the copied dentry is hashed and becomes
1877 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1879 * To prevent any new references to the target during the
1880 * rename, we unhash the dentry in advance.
1882 if (!d_unhashed(new_dentry)) {
1884 rehash = new_dentry;
1887 if (atomic_read(&new_dentry->d_count) > 2) {
1890 /* copy the target dentry's name */
1891 dentry = d_alloc(new_dentry->d_parent,
1892 &new_dentry->d_name);
1896 /* silly-rename the existing target ... */
1897 err = nfs_sillyrename(new_dir, new_dentry);
1901 new_dentry = dentry;
1907 nfs_inode_return_delegation(old_inode);
1908 if (new_inode != NULL)
1909 nfs_inode_return_delegation(new_inode);
1911 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1912 new_dir, &new_dentry->d_name);
1913 nfs_mark_for_revalidate(old_inode);
1918 if (new_inode != NULL)
1919 nfs_drop_nlink(new_inode);
1920 d_move(old_dentry, new_dentry);
1921 nfs_set_verifier(new_dentry,
1922 nfs_save_change_attribute(new_dir));
1923 } else if (error == -ENOENT)
1924 nfs_dentry_handle_enoent(old_dentry);
1926 /* new dentry created? */
1932 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1933 static LIST_HEAD(nfs_access_lru_list);
1934 static atomic_long_t nfs_access_nr_entries;
1936 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1938 put_rpccred(entry->cred);
1940 smp_mb__before_atomic_dec();
1941 atomic_long_dec(&nfs_access_nr_entries);
1942 smp_mb__after_atomic_dec();
1945 static void nfs_access_free_list(struct list_head *head)
1947 struct nfs_access_entry *cache;
1949 while (!list_empty(head)) {
1950 cache = list_entry(head->next, struct nfs_access_entry, lru);
1951 list_del(&cache->lru);
1952 nfs_access_free_entry(cache);
1956 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1959 struct nfs_inode *nfsi, *next;
1960 struct nfs_access_entry *cache;
1962 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1963 return (nr_to_scan == 0) ? 0 : -1;
1965 spin_lock(&nfs_access_lru_lock);
1966 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1967 struct inode *inode;
1969 if (nr_to_scan-- == 0)
1971 inode = &nfsi->vfs_inode;
1972 spin_lock(&inode->i_lock);
1973 if (list_empty(&nfsi->access_cache_entry_lru))
1974 goto remove_lru_entry;
1975 cache = list_entry(nfsi->access_cache_entry_lru.next,
1976 struct nfs_access_entry, lru);
1977 list_move(&cache->lru, &head);
1978 rb_erase(&cache->rb_node, &nfsi->access_cache);
1979 if (!list_empty(&nfsi->access_cache_entry_lru))
1980 list_move_tail(&nfsi->access_cache_inode_lru,
1981 &nfs_access_lru_list);
1984 list_del_init(&nfsi->access_cache_inode_lru);
1985 smp_mb__before_clear_bit();
1986 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1987 smp_mb__after_clear_bit();
1989 spin_unlock(&inode->i_lock);
1991 spin_unlock(&nfs_access_lru_lock);
1992 nfs_access_free_list(&head);
1993 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1996 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1998 struct rb_root *root_node = &nfsi->access_cache;
2000 struct nfs_access_entry *entry;
2002 /* Unhook entries from the cache */
2003 while ((n = rb_first(root_node)) != NULL) {
2004 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2005 rb_erase(n, root_node);
2006 list_move(&entry->lru, head);
2008 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2011 void nfs_access_zap_cache(struct inode *inode)
2015 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2017 /* Remove from global LRU init */
2018 spin_lock(&nfs_access_lru_lock);
2019 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2020 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2022 spin_lock(&inode->i_lock);
2023 __nfs_access_zap_cache(NFS_I(inode), &head);
2024 spin_unlock(&inode->i_lock);
2025 spin_unlock(&nfs_access_lru_lock);
2026 nfs_access_free_list(&head);
2029 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2031 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2032 struct nfs_access_entry *entry;
2035 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2037 if (cred < entry->cred)
2039 else if (cred > entry->cred)
2047 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2049 struct nfs_inode *nfsi = NFS_I(inode);
2050 struct nfs_access_entry *cache;
2053 spin_lock(&inode->i_lock);
2054 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2056 cache = nfs_access_search_rbtree(inode, cred);
2059 if (!nfs_have_delegated_attributes(inode) &&
2060 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2062 res->jiffies = cache->jiffies;
2063 res->cred = cache->cred;
2064 res->mask = cache->mask;
2065 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2068 spin_unlock(&inode->i_lock);
2071 rb_erase(&cache->rb_node, &nfsi->access_cache);
2072 list_del(&cache->lru);
2073 spin_unlock(&inode->i_lock);
2074 nfs_access_free_entry(cache);
2077 spin_unlock(&inode->i_lock);
2078 nfs_access_zap_cache(inode);
2082 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2084 struct nfs_inode *nfsi = NFS_I(inode);
2085 struct rb_root *root_node = &nfsi->access_cache;
2086 struct rb_node **p = &root_node->rb_node;
2087 struct rb_node *parent = NULL;
2088 struct nfs_access_entry *entry;
2090 spin_lock(&inode->i_lock);
2091 while (*p != NULL) {
2093 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2095 if (set->cred < entry->cred)
2096 p = &parent->rb_left;
2097 else if (set->cred > entry->cred)
2098 p = &parent->rb_right;
2102 rb_link_node(&set->rb_node, parent, p);
2103 rb_insert_color(&set->rb_node, root_node);
2104 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2105 spin_unlock(&inode->i_lock);
2108 rb_replace_node(parent, &set->rb_node, root_node);
2109 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2110 list_del(&entry->lru);
2111 spin_unlock(&inode->i_lock);
2112 nfs_access_free_entry(entry);
2115 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2117 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2120 RB_CLEAR_NODE(&cache->rb_node);
2121 cache->jiffies = set->jiffies;
2122 cache->cred = get_rpccred(set->cred);
2123 cache->mask = set->mask;
2125 nfs_access_add_rbtree(inode, cache);
2127 /* Update accounting */
2128 smp_mb__before_atomic_inc();
2129 atomic_long_inc(&nfs_access_nr_entries);
2130 smp_mb__after_atomic_inc();
2132 /* Add inode to global LRU list */
2133 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2134 spin_lock(&nfs_access_lru_lock);
2135 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2136 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2137 &nfs_access_lru_list);
2138 spin_unlock(&nfs_access_lru_lock);
2142 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2144 struct nfs_access_entry cache;
2147 status = nfs_access_get_cached(inode, cred, &cache);
2151 /* Be clever: ask server to check for all possible rights */
2152 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2154 cache.jiffies = jiffies;
2155 status = NFS_PROTO(inode)->access(inode, &cache);
2157 if (status == -ESTALE) {
2158 nfs_zap_caches(inode);
2159 if (!S_ISDIR(inode->i_mode))
2160 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2164 nfs_access_add_cache(inode, &cache);
2166 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2171 static int nfs_open_permission_mask(int openflags)
2175 if (openflags & FMODE_READ)
2177 if (openflags & FMODE_WRITE)
2179 if (openflags & FMODE_EXEC)
2184 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2186 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2189 int nfs_permission(struct inode *inode, int mask)
2191 struct rpc_cred *cred;
2194 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2196 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2198 /* Is this sys_access() ? */
2199 if (mask & (MAY_ACCESS | MAY_CHDIR))
2202 switch (inode->i_mode & S_IFMT) {
2206 /* NFSv4 has atomic_open... */
2207 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2208 && (mask & MAY_OPEN)
2209 && !(mask & MAY_EXEC))
2214 * Optimize away all write operations, since the server
2215 * will check permissions when we perform the op.
2217 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2222 if (!NFS_PROTO(inode)->access)
2225 cred = rpc_lookup_cred();
2226 if (!IS_ERR(cred)) {
2227 res = nfs_do_access(inode, cred, mask);
2230 res = PTR_ERR(cred);
2232 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2235 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2236 inode->i_sb->s_id, inode->i_ino, mask, res);
2239 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2241 res = generic_permission(inode, mask, NULL);
2247 * version-control: t
2248 * kept-new-versions: 5