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>
37 #include <linux/kmemleak.h>
38 #include <linux/xattr.h>
40 #include "delegation.h"
45 /* #define NFS_DEBUG_VERBOSE 1 */
47 static int nfs_opendir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, int);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
63 const struct file_operations nfs_dir_operations = {
64 .llseek = nfs_llseek_dir,
65 .read = generic_read_dir,
66 .readdir = nfs_readdir,
68 .release = nfs_release,
69 .fsync = nfs_fsync_dir,
72 const struct inode_operations nfs_dir_inode_operations = {
77 .symlink = nfs_symlink,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
87 const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
92 const struct inode_operations nfs3_dir_inode_operations = {
97 .symlink = nfs_symlink,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
110 #endif /* CONFIG_NFS_V3 */
114 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
116 const struct inode_operations nfs4_dir_inode_operations = {
117 .create = nfs_open_create,
118 .lookup = nfs_atomic_lookup,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
135 #endif /* CONFIG_NFS_V4 */
141 nfs_opendir(struct inode *inode, struct file *filp)
145 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
146 filp->f_path.dentry->d_parent->d_name.name,
147 filp->f_path.dentry->d_name.name);
149 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
151 /* Call generic open code in order to cache credentials */
152 res = nfs_open(inode, filp);
153 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
154 /* This is a mountpoint, so d_revalidate will never
155 * have been called, so we need to refresh the
156 * inode (for close-open consistency) ourselves.
158 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
163 struct nfs_cache_array_entry {
167 unsigned char d_type;
170 struct nfs_cache_array {
174 struct nfs_cache_array_entry array[0];
177 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
181 unsigned long page_index;
184 loff_t current_index;
185 decode_dirent_t decode;
187 unsigned long timestamp;
188 unsigned long gencount;
189 unsigned int cache_entry_index;
192 } nfs_readdir_descriptor_t;
195 * The caller is responsible for calling nfs_readdir_release_array(page)
198 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
202 return ERR_PTR(-EIO);
205 return ERR_PTR(-ENOMEM);
210 void nfs_readdir_release_array(struct page *page)
216 * we are freeing strings created by nfs_add_to_readdir_array()
219 void nfs_readdir_clear_array(struct page *page)
221 struct nfs_cache_array *array;
224 array = kmap_atomic(page, KM_USER0);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].string.name);
227 kunmap_atomic(array, KM_USER0);
231 * the caller is responsible for freeing qstr.name
232 * when called by nfs_readdir_add_to_array, the strings will be freed in
233 * nfs_clear_readdir_array()
236 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
239 string->name = kmemdup(name, len, GFP_KERNEL);
240 if (string->name == NULL)
243 * Avoid a kmemleak false positive. The pointer to the name is stored
244 * in a page cache page which kmemleak does not scan.
246 kmemleak_not_leak(string->name);
247 string->hash = full_name_hash(name, len);
252 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
254 struct nfs_cache_array *array = nfs_readdir_get_array(page);
255 struct nfs_cache_array_entry *cache_entry;
259 return PTR_ERR(array);
261 cache_entry = &array->array[array->size];
263 /* Check that this entry lies within the page bounds */
265 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
268 cache_entry->cookie = entry->prev_cookie;
269 cache_entry->ino = entry->ino;
270 cache_entry->d_type = entry->d_type;
271 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
274 array->last_cookie = entry->cookie;
277 array->eof_index = array->size;
279 nfs_readdir_release_array(page);
284 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
286 loff_t diff = desc->file->f_pos - desc->current_index;
291 if (diff >= array->size) {
292 if (array->eof_index >= 0)
294 desc->current_index += array->size;
298 index = (unsigned int)diff;
299 *desc->dir_cookie = array->array[index].cookie;
300 desc->cache_entry_index = index;
308 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
311 int status = -EAGAIN;
313 for (i = 0; i < array->size; i++) {
314 if (array->array[i].cookie == *desc->dir_cookie) {
315 desc->cache_entry_index = i;
319 if (array->eof_index >= 0) {
320 status = -EBADCOOKIE;
321 if (*desc->dir_cookie == array->last_cookie)
328 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
330 struct nfs_cache_array *array;
333 array = nfs_readdir_get_array(desc->page);
335 status = PTR_ERR(array);
339 if (*desc->dir_cookie == 0)
340 status = nfs_readdir_search_for_pos(array, desc);
342 status = nfs_readdir_search_for_cookie(array, desc);
344 if (status == -EAGAIN) {
345 desc->last_cookie = array->last_cookie;
348 nfs_readdir_release_array(desc->page);
353 /* Fill a page with xdr information before transferring to the cache page */
355 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
356 struct nfs_entry *entry, struct file *file, struct inode *inode)
358 struct rpc_cred *cred = nfs_file_cred(file);
359 unsigned long timestamp, gencount;
364 gencount = nfs_inc_attr_generation_counter();
365 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
366 NFS_SERVER(inode)->dtsize, desc->plus);
368 /* We requested READDIRPLUS, but the server doesn't grok it */
369 if (error == -ENOTSUPP && desc->plus) {
370 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
371 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
377 desc->timestamp = timestamp;
378 desc->gencount = gencount;
383 static int xdr_decode(nfs_readdir_descriptor_t *desc,
384 struct nfs_entry *entry, struct xdr_stream *xdr)
388 error = desc->decode(xdr, entry, desc->plus);
391 entry->fattr->time_start = desc->timestamp;
392 entry->fattr->gencount = desc->gencount;
397 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
399 if (dentry->d_inode == NULL)
401 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
409 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
411 struct qstr filename = {
415 struct dentry *dentry;
416 struct dentry *alias;
417 struct inode *dir = parent->d_inode;
420 if (filename.name[0] == '.') {
421 if (filename.len == 1)
423 if (filename.len == 2 && filename.name[1] == '.')
426 filename.hash = full_name_hash(filename.name, filename.len);
428 dentry = d_lookup(parent, &filename);
429 if (dentry != NULL) {
430 if (nfs_same_file(dentry, entry)) {
431 nfs_refresh_inode(dentry->d_inode, entry->fattr);
439 dentry = d_alloc(parent, &filename);
443 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
444 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
448 alias = d_materialise_unique(dentry, inode);
452 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
455 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
461 /* Perform conversion from xdr to cache array */
463 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
464 void *xdr_page, struct page *page, unsigned int buflen)
466 struct xdr_stream stream;
468 __be32 *ptr = xdr_page;
469 struct nfs_cache_array *array;
470 unsigned int count = 0;
473 buf.head->iov_base = xdr_page;
474 buf.head->iov_len = buflen;
475 buf.tail->iov_len = 0;
478 buf.buflen = buf.head->iov_len;
479 buf.len = buf.head->iov_len;
481 xdr_init_decode(&stream, &buf, ptr);
485 status = xdr_decode(desc, entry, &stream);
487 if (status == -EAGAIN)
495 nfs_prime_dcache(desc->file->f_path.dentry, entry);
497 status = nfs_readdir_add_to_array(entry, page);
500 } while (!entry->eof);
502 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
503 array = nfs_readdir_get_array(page);
504 if (!IS_ERR(array)) {
505 array->eof_index = array->size;
507 nfs_readdir_release_array(page);
509 status = PTR_ERR(array);
515 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
518 for (i = 0; i < npages; i++)
523 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
526 vm_unmap_ram(ptr, npages);
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 void *nfs_readdir_large_page(struct page **pages, unsigned int npages)
540 for (i = 0; i < npages; i++) {
541 struct page *page = alloc_page(GFP_KERNEL);
547 ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL);
548 if (!IS_ERR_OR_NULL(ptr))
551 nfs_readdir_free_pagearray(pages, i);
556 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
558 struct page *pages[NFS_MAX_READDIR_PAGES];
559 void *pages_ptr = NULL;
560 struct nfs_entry entry;
561 struct file *file = desc->file;
562 struct nfs_cache_array *array;
563 int status = -ENOMEM;
564 unsigned int array_size = ARRAY_SIZE(pages);
566 entry.prev_cookie = 0;
567 entry.cookie = desc->last_cookie;
569 entry.fh = nfs_alloc_fhandle();
570 entry.fattr = nfs_alloc_fattr();
571 entry.server = NFS_SERVER(inode);
572 if (entry.fh == NULL || entry.fattr == NULL)
575 array = nfs_readdir_get_array(page);
577 status = PTR_ERR(array);
580 memset(array, 0, sizeof(struct nfs_cache_array));
581 array->eof_index = -1;
583 pages_ptr = nfs_readdir_large_page(pages, array_size);
585 goto out_release_array;
588 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
593 status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen);
595 if (status == -ENOSPC)
599 } while (array->eof_index < 0);
601 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
603 nfs_readdir_release_array(page);
605 nfs_free_fattr(entry.fattr);
606 nfs_free_fhandle(entry.fh);
611 * Now we cache directories properly, by converting xdr information
612 * to an array that can be used for lookups later. This results in
613 * fewer cache pages, since we can store more information on each page.
614 * We only need to convert from xdr once so future lookups are much simpler
617 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
619 struct inode *inode = desc->file->f_path.dentry->d_inode;
622 ret = nfs_readdir_xdr_to_array(desc, page, inode);
625 SetPageUptodate(page);
627 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
628 /* Should never happen */
629 nfs_zap_mapping(inode, inode->i_mapping);
639 void cache_page_release(nfs_readdir_descriptor_t *desc)
641 if (!desc->page->mapping)
642 nfs_readdir_clear_array(desc->page);
643 page_cache_release(desc->page);
648 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
650 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
651 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
655 * Returns 0 if desc->dir_cookie was found on page desc->page_index
658 int find_cache_page(nfs_readdir_descriptor_t *desc)
662 desc->page = get_cache_page(desc);
663 if (IS_ERR(desc->page))
664 return PTR_ERR(desc->page);
666 res = nfs_readdir_search_array(desc);
668 cache_page_release(desc);
672 /* Search for desc->dir_cookie from the beginning of the page cache */
674 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
678 if (desc->page_index == 0) {
679 desc->current_index = 0;
680 desc->last_cookie = 0;
683 res = find_cache_page(desc);
684 } while (res == -EAGAIN);
689 * Once we've found the start of the dirent within a page: fill 'er up...
692 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
695 struct file *file = desc->file;
698 struct nfs_cache_array *array = NULL;
700 array = nfs_readdir_get_array(desc->page);
702 res = PTR_ERR(array);
706 for (i = desc->cache_entry_index; i < array->size; i++) {
707 struct nfs_cache_array_entry *ent;
709 ent = &array->array[i];
710 if (filldir(dirent, ent->string.name, ent->string.len,
711 file->f_pos, nfs_compat_user_ino64(ent->ino),
717 if (i < (array->size-1))
718 *desc->dir_cookie = array->array[i+1].cookie;
720 *desc->dir_cookie = array->last_cookie;
722 if (array->eof_index >= 0)
725 nfs_readdir_release_array(desc->page);
727 cache_page_release(desc);
728 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
729 (unsigned long long)*desc->dir_cookie, res);
734 * If we cannot find a cookie in our cache, we suspect that this is
735 * because it points to a deleted file, so we ask the server to return
736 * whatever it thinks is the next entry. We then feed this to filldir.
737 * If all goes well, we should then be able to find our way round the
738 * cache on the next call to readdir_search_pagecache();
740 * NOTE: we cannot add the anonymous page to the pagecache because
741 * the data it contains might not be page aligned. Besides,
742 * we should already have a complete representation of the
743 * directory in the page cache by the time we get here.
746 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
749 struct page *page = NULL;
751 struct inode *inode = desc->file->f_path.dentry->d_inode;
753 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
754 (unsigned long long)*desc->dir_cookie);
756 page = alloc_page(GFP_HIGHUSER);
762 desc->page_index = 0;
763 desc->last_cookie = *desc->dir_cookie;
766 status = nfs_readdir_xdr_to_array(desc, page, inode);
770 status = nfs_do_filldir(desc, dirent, filldir);
773 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
777 cache_page_release(desc);
781 /* The file offset position represents the dirent entry number. A
782 last cookie cache takes care of the common case of reading the
785 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
787 struct dentry *dentry = filp->f_path.dentry;
788 struct inode *inode = dentry->d_inode;
789 nfs_readdir_descriptor_t my_desc,
793 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
794 dentry->d_parent->d_name.name, dentry->d_name.name,
795 (long long)filp->f_pos);
796 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
799 * filp->f_pos points to the dirent entry number.
800 * *desc->dir_cookie has the cookie for the next entry. We have
801 * to either find the entry with the appropriate number or
802 * revalidate the cookie.
804 memset(desc, 0, sizeof(*desc));
807 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
808 desc->decode = NFS_PROTO(inode)->decode_dirent;
809 desc->plus = NFS_USE_READDIRPLUS(inode);
811 nfs_block_sillyrename(dentry);
812 res = nfs_revalidate_mapping(inode, filp->f_mapping);
817 res = readdir_search_pagecache(desc);
819 if (res == -EBADCOOKIE) {
821 /* This means either end of directory */
822 if (*desc->dir_cookie && desc->eof == 0) {
823 /* Or that the server has 'lost' a cookie */
824 res = uncached_readdir(desc, dirent, filldir);
830 if (res == -ETOOSMALL && desc->plus) {
831 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
832 nfs_zap_caches(inode);
833 desc->page_index = 0;
841 res = nfs_do_filldir(desc, dirent, filldir);
844 } while (!desc->eof);
846 nfs_unblock_sillyrename(dentry);
849 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
850 dentry->d_parent->d_name.name, dentry->d_name.name,
855 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
857 struct dentry *dentry = filp->f_path.dentry;
858 struct inode *inode = dentry->d_inode;
860 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
861 dentry->d_parent->d_name.name,
865 mutex_lock(&inode->i_mutex);
868 offset += filp->f_pos;
876 if (offset != filp->f_pos) {
877 filp->f_pos = offset;
878 nfs_file_open_context(filp)->dir_cookie = 0;
881 mutex_unlock(&inode->i_mutex);
886 * All directory operations under NFS are synchronous, so fsync()
887 * is a dummy operation.
889 static int nfs_fsync_dir(struct file *filp, int datasync)
891 struct dentry *dentry = filp->f_path.dentry;
893 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
894 dentry->d_parent->d_name.name, dentry->d_name.name,
897 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
902 * nfs_force_lookup_revalidate - Mark the directory as having changed
903 * @dir - pointer to directory inode
905 * This forces the revalidation code in nfs_lookup_revalidate() to do a
906 * full lookup on all child dentries of 'dir' whenever a change occurs
907 * on the server that might have invalidated our dcache.
909 * The caller should be holding dir->i_lock
911 void nfs_force_lookup_revalidate(struct inode *dir)
913 NFS_I(dir)->cache_change_attribute++;
917 * A check for whether or not the parent directory has changed.
918 * In the case it has, we assume that the dentries are untrustworthy
919 * and may need to be looked up again.
921 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
925 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
927 if (!nfs_verify_change_attribute(dir, dentry->d_time))
929 /* Revalidate nfsi->cache_change_attribute before we declare a match */
930 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
932 if (!nfs_verify_change_attribute(dir, dentry->d_time))
938 * Return the intent data that applies to this particular path component
940 * Note that the current set of intents only apply to the very last
941 * component of the path.
942 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
944 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
946 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
948 return nd->flags & mask;
952 * Use intent information to check whether or not we're going to do
953 * an O_EXCL create using this path component.
955 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
957 if (NFS_PROTO(dir)->version == 2)
959 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
963 * Inode and filehandle revalidation for lookups.
965 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
966 * or if the intent information indicates that we're about to open this
967 * particular file and the "nocto" mount flag is not set.
971 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
973 struct nfs_server *server = NFS_SERVER(inode);
975 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
978 /* VFS wants an on-the-wire revalidation */
979 if (nd->flags & LOOKUP_REVAL)
981 /* This is an open(2) */
982 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
983 !(server->flags & NFS_MOUNT_NOCTO) &&
984 (S_ISREG(inode->i_mode) ||
985 S_ISDIR(inode->i_mode)))
989 return nfs_revalidate_inode(server, inode);
991 return __nfs_revalidate_inode(server, inode);
995 * We judge how long we want to trust negative
996 * dentries by looking at the parent inode mtime.
998 * If parent mtime has changed, we revalidate, else we wait for a
999 * period corresponding to the parent's attribute cache timeout value.
1002 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1003 struct nameidata *nd)
1005 /* Don't revalidate a negative dentry if we're creating a new file */
1006 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1008 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1010 return !nfs_check_verifier(dir, dentry);
1014 * This is called every time the dcache has a lookup hit,
1015 * and we should check whether we can really trust that
1018 * NOTE! The hit can be a negative hit too, don't assume
1021 * If the parent directory is seen to have changed, we throw out the
1022 * cached dentry and do a new lookup.
1024 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
1027 struct inode *inode;
1028 struct dentry *parent;
1029 struct nfs_fh *fhandle = NULL;
1030 struct nfs_fattr *fattr = NULL;
1033 parent = dget_parent(dentry);
1034 dir = parent->d_inode;
1035 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1036 inode = dentry->d_inode;
1039 if (nfs_neg_need_reval(dir, dentry, nd))
1044 if (is_bad_inode(inode)) {
1045 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1046 __func__, dentry->d_parent->d_name.name,
1047 dentry->d_name.name);
1051 if (nfs_have_delegation(inode, FMODE_READ))
1052 goto out_set_verifier;
1054 /* Force a full look up iff the parent directory has changed */
1055 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1056 if (nfs_lookup_verify_inode(inode, nd))
1057 goto out_zap_parent;
1061 if (NFS_STALE(inode))
1065 fhandle = nfs_alloc_fhandle();
1066 fattr = nfs_alloc_fattr();
1067 if (fhandle == NULL || fattr == NULL)
1070 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1073 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1075 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1078 nfs_free_fattr(fattr);
1079 nfs_free_fhandle(fhandle);
1081 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1084 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1085 __func__, dentry->d_parent->d_name.name,
1086 dentry->d_name.name);
1089 nfs_zap_caches(dir);
1091 nfs_mark_for_revalidate(dir);
1092 if (inode && S_ISDIR(inode->i_mode)) {
1093 /* Purge readdir caches. */
1094 nfs_zap_caches(inode);
1095 /* If we have submounts, don't unhash ! */
1096 if (have_submounts(dentry))
1098 if (dentry->d_flags & DCACHE_DISCONNECTED)
1100 shrink_dcache_parent(dentry);
1103 nfs_free_fattr(fattr);
1104 nfs_free_fhandle(fhandle);
1106 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1107 __func__, dentry->d_parent->d_name.name,
1108 dentry->d_name.name);
1111 nfs_free_fattr(fattr);
1112 nfs_free_fhandle(fhandle);
1114 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1115 __func__, dentry->d_parent->d_name.name,
1116 dentry->d_name.name, error);
1121 * This is called from dput() when d_count is going to 0.
1123 static int nfs_dentry_delete(struct dentry *dentry)
1125 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1126 dentry->d_parent->d_name.name, dentry->d_name.name,
1129 /* Unhash any dentry with a stale inode */
1130 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1133 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1134 /* Unhash it, so that ->d_iput() would be called */
1137 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1138 /* Unhash it, so that ancestors of killed async unlink
1139 * files will be cleaned up during umount */
1146 static void nfs_drop_nlink(struct inode *inode)
1148 spin_lock(&inode->i_lock);
1149 if (inode->i_nlink > 0)
1151 spin_unlock(&inode->i_lock);
1155 * Called when the dentry loses inode.
1156 * We use it to clean up silly-renamed files.
1158 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1160 if (S_ISDIR(inode->i_mode))
1161 /* drop any readdir cache as it could easily be old */
1162 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1164 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1166 nfs_complete_unlink(dentry, inode);
1171 const struct dentry_operations nfs_dentry_operations = {
1172 .d_revalidate = nfs_lookup_revalidate,
1173 .d_delete = nfs_dentry_delete,
1174 .d_iput = nfs_dentry_iput,
1177 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1180 struct dentry *parent;
1181 struct inode *inode = NULL;
1182 struct nfs_fh *fhandle = NULL;
1183 struct nfs_fattr *fattr = NULL;
1186 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1187 dentry->d_parent->d_name.name, dentry->d_name.name);
1188 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1190 res = ERR_PTR(-ENAMETOOLONG);
1191 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1194 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1197 * If we're doing an exclusive create, optimize away the lookup
1198 * but don't hash the dentry.
1200 if (nfs_is_exclusive_create(dir, nd)) {
1201 d_instantiate(dentry, NULL);
1206 res = ERR_PTR(-ENOMEM);
1207 fhandle = nfs_alloc_fhandle();
1208 fattr = nfs_alloc_fattr();
1209 if (fhandle == NULL || fattr == NULL)
1212 parent = dentry->d_parent;
1213 /* Protect against concurrent sillydeletes */
1214 nfs_block_sillyrename(parent);
1215 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1216 if (error == -ENOENT)
1219 res = ERR_PTR(error);
1220 goto out_unblock_sillyrename;
1222 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1223 res = ERR_CAST(inode);
1225 goto out_unblock_sillyrename;
1228 res = d_materialise_unique(dentry, inode);
1231 goto out_unblock_sillyrename;
1234 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1235 out_unblock_sillyrename:
1236 nfs_unblock_sillyrename(parent);
1238 nfs_free_fattr(fattr);
1239 nfs_free_fhandle(fhandle);
1243 #ifdef CONFIG_NFS_V4
1244 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1246 const struct dentry_operations nfs4_dentry_operations = {
1247 .d_revalidate = nfs_open_revalidate,
1248 .d_delete = nfs_dentry_delete,
1249 .d_iput = nfs_dentry_iput,
1253 * Use intent information to determine whether we need to substitute
1254 * the NFSv4-style stateful OPEN for the LOOKUP call
1256 static int is_atomic_open(struct nameidata *nd)
1258 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1260 /* NFS does not (yet) have a stateful open for directories */
1261 if (nd->flags & LOOKUP_DIRECTORY)
1263 /* Are we trying to write to a read only partition? */
1264 if (__mnt_is_readonly(nd->path.mnt) &&
1265 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1270 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1272 struct path path = {
1273 .mnt = nd->path.mnt,
1276 struct nfs_open_context *ctx;
1277 struct rpc_cred *cred;
1278 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1280 cred = rpc_lookup_cred();
1282 return ERR_CAST(cred);
1283 ctx = alloc_nfs_open_context(&path, cred, fmode);
1286 return ERR_PTR(-ENOMEM);
1290 static int do_open(struct inode *inode, struct file *filp)
1292 nfs_fscache_set_inode_cookie(inode, filp);
1296 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1301 /* If the open_intent is for execute, we have an extra check to make */
1302 if (ctx->mode & FMODE_EXEC) {
1303 ret = nfs_may_open(ctx->path.dentry->d_inode,
1305 nd->intent.open.flags);
1309 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1311 ret = PTR_ERR(filp);
1313 nfs_file_set_open_context(filp, ctx);
1315 put_nfs_open_context(ctx);
1319 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1321 struct nfs_open_context *ctx;
1323 struct dentry *res = NULL;
1324 struct inode *inode;
1328 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1329 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1331 /* Check that we are indeed trying to open this file */
1332 if (!is_atomic_open(nd))
1335 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1336 res = ERR_PTR(-ENAMETOOLONG);
1339 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1341 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1343 if (nd->flags & LOOKUP_EXCL) {
1344 d_instantiate(dentry, NULL);
1348 ctx = nameidata_to_nfs_open_context(dentry, nd);
1349 res = ERR_CAST(ctx);
1353 open_flags = nd->intent.open.flags;
1354 if (nd->flags & LOOKUP_CREATE) {
1355 attr.ia_mode = nd->intent.open.create_mode;
1356 attr.ia_valid = ATTR_MODE;
1357 attr.ia_mode &= ~current_umask();
1359 open_flags &= ~(O_EXCL | O_CREAT);
1363 /* Open the file on the server */
1364 nfs_block_sillyrename(dentry->d_parent);
1365 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1366 if (IS_ERR(inode)) {
1367 nfs_unblock_sillyrename(dentry->d_parent);
1368 put_nfs_open_context(ctx);
1369 switch (PTR_ERR(inode)) {
1370 /* Make a negative dentry */
1372 d_add(dentry, NULL);
1375 /* This turned out not to be a regular file */
1379 if (!(nd->intent.open.flags & O_NOFOLLOW))
1384 res = ERR_CAST(inode);
1388 res = d_add_unique(dentry, inode);
1389 nfs_unblock_sillyrename(dentry->d_parent);
1391 dput(ctx->path.dentry);
1392 ctx->path.dentry = dget(res);
1395 err = nfs_intent_set_file(nd, ctx);
1399 return ERR_PTR(err);
1402 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1405 return nfs_lookup(dir, dentry, nd);
1408 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1410 struct dentry *parent = NULL;
1411 struct inode *inode = dentry->d_inode;
1413 struct nfs_open_context *ctx;
1414 int openflags, ret = 0;
1416 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1419 parent = dget_parent(dentry);
1420 dir = parent->d_inode;
1422 /* We can't create new files in nfs_open_revalidate(), so we
1423 * optimize away revalidation of negative dentries.
1425 if (inode == NULL) {
1426 if (!nfs_neg_need_reval(dir, dentry, nd))
1431 /* NFS only supports OPEN on regular files */
1432 if (!S_ISREG(inode->i_mode))
1434 openflags = nd->intent.open.flags;
1435 /* We cannot do exclusive creation on a positive dentry */
1436 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1438 /* We can't create new files, or truncate existing ones here */
1439 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1441 ctx = nameidata_to_nfs_open_context(dentry, nd);
1446 * Note: we're not holding inode->i_mutex and so may be racing with
1447 * operations that change the directory. We therefore save the
1448 * change attribute *before* we do the RPC call.
1450 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1451 if (IS_ERR(inode)) {
1452 ret = PTR_ERR(inode);
1465 if (inode != dentry->d_inode)
1468 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1469 ret = nfs_intent_set_file(nd, ctx);
1479 put_nfs_open_context(ctx);
1485 return nfs_lookup_revalidate(dentry, nd);
1488 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1489 struct nameidata *nd)
1491 struct nfs_open_context *ctx = NULL;
1496 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1497 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1499 attr.ia_mode = mode;
1500 attr.ia_valid = ATTR_MODE;
1502 if ((nd->flags & LOOKUP_CREATE) != 0) {
1503 open_flags = nd->intent.open.flags;
1505 ctx = nameidata_to_nfs_open_context(dentry, nd);
1506 error = PTR_ERR(ctx);
1511 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1515 error = nfs_intent_set_file(nd, ctx);
1522 put_nfs_open_context(ctx);
1529 #endif /* CONFIG_NFSV4 */
1532 * Code common to create, mkdir, and mknod.
1534 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1535 struct nfs_fattr *fattr)
1537 struct dentry *parent = dget_parent(dentry);
1538 struct inode *dir = parent->d_inode;
1539 struct inode *inode;
1540 int error = -EACCES;
1544 /* We may have been initialized further down */
1545 if (dentry->d_inode)
1547 if (fhandle->size == 0) {
1548 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1552 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1553 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1554 struct nfs_server *server = NFS_SB(dentry->d_sb);
1555 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1559 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1560 error = PTR_ERR(inode);
1563 d_add(dentry, inode);
1568 nfs_mark_for_revalidate(dir);
1574 * Following a failed create operation, we drop the dentry rather
1575 * than retain a negative dentry. This avoids a problem in the event
1576 * that the operation succeeded on the server, but an error in the
1577 * reply path made it appear to have failed.
1579 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1580 struct nameidata *nd)
1585 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1586 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1588 attr.ia_mode = mode;
1589 attr.ia_valid = ATTR_MODE;
1591 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1601 * See comments for nfs_proc_create regarding failed operations.
1604 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1609 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1610 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1612 if (!new_valid_dev(rdev))
1615 attr.ia_mode = mode;
1616 attr.ia_valid = ATTR_MODE;
1618 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1628 * See comments for nfs_proc_create regarding failed operations.
1630 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1635 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1636 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1638 attr.ia_valid = ATTR_MODE;
1639 attr.ia_mode = mode | S_IFDIR;
1641 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1650 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1652 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1656 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1660 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1661 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1663 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1664 /* Ensure the VFS deletes this inode */
1665 if (error == 0 && dentry->d_inode != NULL)
1666 clear_nlink(dentry->d_inode);
1667 else if (error == -ENOENT)
1668 nfs_dentry_handle_enoent(dentry);
1674 * Remove a file after making sure there are no pending writes,
1675 * and after checking that the file has only one user.
1677 * We invalidate the attribute cache and free the inode prior to the operation
1678 * to avoid possible races if the server reuses the inode.
1680 static int nfs_safe_remove(struct dentry *dentry)
1682 struct inode *dir = dentry->d_parent->d_inode;
1683 struct inode *inode = dentry->d_inode;
1686 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1687 dentry->d_parent->d_name.name, dentry->d_name.name);
1689 /* If the dentry was sillyrenamed, we simply call d_delete() */
1690 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1695 if (inode != NULL) {
1696 nfs_inode_return_delegation(inode);
1697 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1698 /* The VFS may want to delete this inode */
1700 nfs_drop_nlink(inode);
1701 nfs_mark_for_revalidate(inode);
1703 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1704 if (error == -ENOENT)
1705 nfs_dentry_handle_enoent(dentry);
1710 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1711 * belongs to an active ".nfs..." file and we return -EBUSY.
1713 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1715 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1718 int need_rehash = 0;
1720 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1721 dir->i_ino, dentry->d_name.name);
1723 spin_lock(&dcache_lock);
1724 spin_lock(&dentry->d_lock);
1725 if (atomic_read(&dentry->d_count) > 1) {
1726 spin_unlock(&dentry->d_lock);
1727 spin_unlock(&dcache_lock);
1728 /* Start asynchronous writeout of the inode */
1729 write_inode_now(dentry->d_inode, 0);
1730 error = nfs_sillyrename(dir, dentry);
1733 if (!d_unhashed(dentry)) {
1737 spin_unlock(&dentry->d_lock);
1738 spin_unlock(&dcache_lock);
1739 error = nfs_safe_remove(dentry);
1740 if (!error || error == -ENOENT) {
1741 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1742 } else if (need_rehash)
1748 * To create a symbolic link, most file systems instantiate a new inode,
1749 * add a page to it containing the path, then write it out to the disk
1750 * using prepare_write/commit_write.
1752 * Unfortunately the NFS client can't create the in-core inode first
1753 * because it needs a file handle to create an in-core inode (see
1754 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1755 * symlink request has completed on the server.
1757 * So instead we allocate a raw page, copy the symname into it, then do
1758 * the SYMLINK request with the page as the buffer. If it succeeds, we
1759 * now have a new file handle and can instantiate an in-core NFS inode
1760 * and move the raw page into its mapping.
1762 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1764 struct pagevec lru_pvec;
1768 unsigned int pathlen = strlen(symname);
1771 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1772 dir->i_ino, dentry->d_name.name, symname);
1774 if (pathlen > PAGE_SIZE)
1775 return -ENAMETOOLONG;
1777 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1778 attr.ia_valid = ATTR_MODE;
1780 page = alloc_page(GFP_HIGHUSER);
1784 kaddr = kmap_atomic(page, KM_USER0);
1785 memcpy(kaddr, symname, pathlen);
1786 if (pathlen < PAGE_SIZE)
1787 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1788 kunmap_atomic(kaddr, KM_USER0);
1790 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1792 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1793 dir->i_sb->s_id, dir->i_ino,
1794 dentry->d_name.name, symname, error);
1801 * No big deal if we can't add this page to the page cache here.
1802 * READLINK will get the missing page from the server if needed.
1804 pagevec_init(&lru_pvec, 0);
1805 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1807 pagevec_add(&lru_pvec, page);
1808 pagevec_lru_add_file(&lru_pvec);
1809 SetPageUptodate(page);
1818 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1820 struct inode *inode = old_dentry->d_inode;
1823 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1824 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1825 dentry->d_parent->d_name.name, dentry->d_name.name);
1827 nfs_inode_return_delegation(inode);
1830 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1833 d_add(dentry, inode);
1840 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1841 * different file handle for the same inode after a rename (e.g. when
1842 * moving to a different directory). A fail-safe method to do so would
1843 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1844 * rename the old file using the sillyrename stuff. This way, the original
1845 * file in old_dir will go away when the last process iput()s the inode.
1849 * It actually works quite well. One needs to have the possibility for
1850 * at least one ".nfs..." file in each directory the file ever gets
1851 * moved or linked to which happens automagically with the new
1852 * implementation that only depends on the dcache stuff instead of
1853 * using the inode layer
1855 * Unfortunately, things are a little more complicated than indicated
1856 * above. For a cross-directory move, we want to make sure we can get
1857 * rid of the old inode after the operation. This means there must be
1858 * no pending writes (if it's a file), and the use count must be 1.
1859 * If these conditions are met, we can drop the dentries before doing
1862 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1863 struct inode *new_dir, struct dentry *new_dentry)
1865 struct inode *old_inode = old_dentry->d_inode;
1866 struct inode *new_inode = new_dentry->d_inode;
1867 struct dentry *dentry = NULL, *rehash = NULL;
1870 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1871 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1872 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1873 atomic_read(&new_dentry->d_count));
1876 * For non-directories, check whether the target is busy and if so,
1877 * make a copy of the dentry and then do a silly-rename. If the
1878 * silly-rename succeeds, the copied dentry is hashed and becomes
1881 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1883 * To prevent any new references to the target during the
1884 * rename, we unhash the dentry in advance.
1886 if (!d_unhashed(new_dentry)) {
1888 rehash = new_dentry;
1891 if (atomic_read(&new_dentry->d_count) > 2) {
1894 /* copy the target dentry's name */
1895 dentry = d_alloc(new_dentry->d_parent,
1896 &new_dentry->d_name);
1900 /* silly-rename the existing target ... */
1901 err = nfs_sillyrename(new_dir, new_dentry);
1905 new_dentry = dentry;
1911 nfs_inode_return_delegation(old_inode);
1912 if (new_inode != NULL)
1913 nfs_inode_return_delegation(new_inode);
1915 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1916 new_dir, &new_dentry->d_name);
1917 nfs_mark_for_revalidate(old_inode);
1922 if (new_inode != NULL)
1923 nfs_drop_nlink(new_inode);
1924 d_move(old_dentry, new_dentry);
1925 nfs_set_verifier(new_dentry,
1926 nfs_save_change_attribute(new_dir));
1927 } else if (error == -ENOENT)
1928 nfs_dentry_handle_enoent(old_dentry);
1930 /* new dentry created? */
1936 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1937 static LIST_HEAD(nfs_access_lru_list);
1938 static atomic_long_t nfs_access_nr_entries;
1940 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1942 put_rpccred(entry->cred);
1944 smp_mb__before_atomic_dec();
1945 atomic_long_dec(&nfs_access_nr_entries);
1946 smp_mb__after_atomic_dec();
1949 static void nfs_access_free_list(struct list_head *head)
1951 struct nfs_access_entry *cache;
1953 while (!list_empty(head)) {
1954 cache = list_entry(head->next, struct nfs_access_entry, lru);
1955 list_del(&cache->lru);
1956 nfs_access_free_entry(cache);
1960 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1963 struct nfs_inode *nfsi, *next;
1964 struct nfs_access_entry *cache;
1966 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1967 return (nr_to_scan == 0) ? 0 : -1;
1969 spin_lock(&nfs_access_lru_lock);
1970 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1971 struct inode *inode;
1973 if (nr_to_scan-- == 0)
1975 inode = &nfsi->vfs_inode;
1976 spin_lock(&inode->i_lock);
1977 if (list_empty(&nfsi->access_cache_entry_lru))
1978 goto remove_lru_entry;
1979 cache = list_entry(nfsi->access_cache_entry_lru.next,
1980 struct nfs_access_entry, lru);
1981 list_move(&cache->lru, &head);
1982 rb_erase(&cache->rb_node, &nfsi->access_cache);
1983 if (!list_empty(&nfsi->access_cache_entry_lru))
1984 list_move_tail(&nfsi->access_cache_inode_lru,
1985 &nfs_access_lru_list);
1988 list_del_init(&nfsi->access_cache_inode_lru);
1989 smp_mb__before_clear_bit();
1990 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1991 smp_mb__after_clear_bit();
1993 spin_unlock(&inode->i_lock);
1995 spin_unlock(&nfs_access_lru_lock);
1996 nfs_access_free_list(&head);
1997 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2000 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2002 struct rb_root *root_node = &nfsi->access_cache;
2004 struct nfs_access_entry *entry;
2006 /* Unhook entries from the cache */
2007 while ((n = rb_first(root_node)) != NULL) {
2008 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2009 rb_erase(n, root_node);
2010 list_move(&entry->lru, head);
2012 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2015 void nfs_access_zap_cache(struct inode *inode)
2019 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2021 /* Remove from global LRU init */
2022 spin_lock(&nfs_access_lru_lock);
2023 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2024 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2026 spin_lock(&inode->i_lock);
2027 __nfs_access_zap_cache(NFS_I(inode), &head);
2028 spin_unlock(&inode->i_lock);
2029 spin_unlock(&nfs_access_lru_lock);
2030 nfs_access_free_list(&head);
2033 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2035 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2036 struct nfs_access_entry *entry;
2039 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2041 if (cred < entry->cred)
2043 else if (cred > entry->cred)
2051 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2053 struct nfs_inode *nfsi = NFS_I(inode);
2054 struct nfs_access_entry *cache;
2057 spin_lock(&inode->i_lock);
2058 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2060 cache = nfs_access_search_rbtree(inode, cred);
2063 if (!nfs_have_delegated_attributes(inode) &&
2064 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2066 res->jiffies = cache->jiffies;
2067 res->cred = cache->cred;
2068 res->mask = cache->mask;
2069 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2072 spin_unlock(&inode->i_lock);
2075 rb_erase(&cache->rb_node, &nfsi->access_cache);
2076 list_del(&cache->lru);
2077 spin_unlock(&inode->i_lock);
2078 nfs_access_free_entry(cache);
2081 spin_unlock(&inode->i_lock);
2082 nfs_access_zap_cache(inode);
2086 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2088 struct nfs_inode *nfsi = NFS_I(inode);
2089 struct rb_root *root_node = &nfsi->access_cache;
2090 struct rb_node **p = &root_node->rb_node;
2091 struct rb_node *parent = NULL;
2092 struct nfs_access_entry *entry;
2094 spin_lock(&inode->i_lock);
2095 while (*p != NULL) {
2097 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2099 if (set->cred < entry->cred)
2100 p = &parent->rb_left;
2101 else if (set->cred > entry->cred)
2102 p = &parent->rb_right;
2106 rb_link_node(&set->rb_node, parent, p);
2107 rb_insert_color(&set->rb_node, root_node);
2108 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2109 spin_unlock(&inode->i_lock);
2112 rb_replace_node(parent, &set->rb_node, root_node);
2113 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2114 list_del(&entry->lru);
2115 spin_unlock(&inode->i_lock);
2116 nfs_access_free_entry(entry);
2119 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2121 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2124 RB_CLEAR_NODE(&cache->rb_node);
2125 cache->jiffies = set->jiffies;
2126 cache->cred = get_rpccred(set->cred);
2127 cache->mask = set->mask;
2129 nfs_access_add_rbtree(inode, cache);
2131 /* Update accounting */
2132 smp_mb__before_atomic_inc();
2133 atomic_long_inc(&nfs_access_nr_entries);
2134 smp_mb__after_atomic_inc();
2136 /* Add inode to global LRU list */
2137 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2138 spin_lock(&nfs_access_lru_lock);
2139 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2140 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2141 &nfs_access_lru_list);
2142 spin_unlock(&nfs_access_lru_lock);
2146 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2148 struct nfs_access_entry cache;
2151 status = nfs_access_get_cached(inode, cred, &cache);
2155 /* Be clever: ask server to check for all possible rights */
2156 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2158 cache.jiffies = jiffies;
2159 status = NFS_PROTO(inode)->access(inode, &cache);
2161 if (status == -ESTALE) {
2162 nfs_zap_caches(inode);
2163 if (!S_ISDIR(inode->i_mode))
2164 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2168 nfs_access_add_cache(inode, &cache);
2170 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2175 static int nfs_open_permission_mask(int openflags)
2179 if (openflags & FMODE_READ)
2181 if (openflags & FMODE_WRITE)
2183 if (openflags & FMODE_EXEC)
2188 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2190 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2193 int nfs_permission(struct inode *inode, int mask)
2195 struct rpc_cred *cred;
2198 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2200 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2202 /* Is this sys_access() ? */
2203 if (mask & (MAY_ACCESS | MAY_CHDIR))
2206 switch (inode->i_mode & S_IFMT) {
2210 /* NFSv4 has atomic_open... */
2211 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2212 && (mask & MAY_OPEN)
2213 && !(mask & MAY_EXEC))
2218 * Optimize away all write operations, since the server
2219 * will check permissions when we perform the op.
2221 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2226 if (!NFS_PROTO(inode)->access)
2229 cred = rpc_lookup_cred();
2230 if (!IS_ERR(cred)) {
2231 res = nfs_do_access(inode, cred, mask);
2234 res = PTR_ERR(cred);
2236 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2239 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2240 inode->i_sb->s_id, inode->i_ino, mask, res);
2243 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2245 res = generic_permission(inode, mask, NULL);
2251 * version-control: t
2252 * kept-new-versions: 5