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>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_readdir(struct file *, void *, filldir_t);
48 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
49 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
50 static int nfs_mkdir(struct inode *, struct dentry *, int);
51 static int nfs_rmdir(struct inode *, struct dentry *);
52 static int nfs_unlink(struct inode *, struct dentry *);
53 static int nfs_symlink(struct inode *, struct dentry *, const char *);
54 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
55 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
56 static int nfs_rename(struct inode *, struct dentry *,
57 struct inode *, struct dentry *);
58 static int nfs_fsync_dir(struct file *, int);
59 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 static int nfs_readdir_clear_array(struct page*, gfp_t);
62 const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .readdir = nfs_readdir,
67 .release = nfs_release,
68 .fsync = nfs_fsync_dir,
71 const struct inode_operations nfs_dir_inode_operations = {
76 .symlink = nfs_symlink,
81 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
86 const struct address_space_operations nfs_dir_addr_space_ops = {
87 .releasepage = nfs_readdir_clear_array,
91 const struct inode_operations nfs3_dir_inode_operations = {
96 .symlink = nfs_symlink,
100 .rename = nfs_rename,
101 .permission = nfs_permission,
102 .getattr = nfs_getattr,
103 .setattr = nfs_setattr,
104 .listxattr = nfs3_listxattr,
105 .getxattr = nfs3_getxattr,
106 .setxattr = nfs3_setxattr,
107 .removexattr = nfs3_removexattr,
109 #endif /* CONFIG_NFS_V3 */
113 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
114 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
115 const struct inode_operations nfs4_dir_inode_operations = {
116 .create = nfs_open_create,
117 .lookup = nfs_atomic_lookup,
119 .unlink = nfs_unlink,
120 .symlink = nfs_symlink,
124 .rename = nfs_rename,
125 .permission = nfs_permission,
126 .getattr = nfs_getattr,
127 .setattr = nfs_setattr,
128 .getxattr = nfs4_getxattr,
129 .setxattr = nfs4_setxattr,
130 .listxattr = nfs4_listxattr,
133 #endif /* CONFIG_NFS_V4 */
139 nfs_opendir(struct inode *inode, struct file *filp)
143 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
144 filp->f_path.dentry->d_parent->d_name.name,
145 filp->f_path.dentry->d_name.name);
147 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
149 /* Call generic open code in order to cache credentials */
150 res = nfs_open(inode, filp);
151 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
152 /* This is a mountpoint, so d_revalidate will never
153 * have been called, so we need to refresh the
154 * inode (for close-open consistency) ourselves.
156 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
161 struct nfs_cache_array_entry {
167 struct nfs_cache_array {
171 struct nfs_cache_array_entry array[0];
174 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
176 typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int);
180 unsigned long page_index;
182 loff_t current_index;
183 decode_dirent_t decode;
185 unsigned long timestamp;
186 unsigned long gencount;
187 unsigned int cache_entry_index;
190 } nfs_readdir_descriptor_t;
193 * The caller is responsible for calling nfs_readdir_release_array(page)
196 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
200 return ERR_PTR(-EIO);
203 return ERR_PTR(-ENOMEM);
208 void nfs_readdir_release_array(struct page *page)
214 * we are freeing strings created by nfs_add_to_readdir_array()
217 int nfs_readdir_clear_array(struct page *page, gfp_t mask)
219 struct nfs_cache_array *array = nfs_readdir_get_array(page);
223 return PTR_ERR(array);
224 for (i = 0; i < array->size; i++)
225 kfree(array->array[i].string.name);
226 nfs_readdir_release_array(page);
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 if (array->size >= MAX_READDIR_ARRAY)
264 cache_entry = &array->array[array->size];
265 cache_entry->cookie = entry->prev_cookie;
266 cache_entry->ino = entry->ino;
267 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
270 array->last_cookie = entry->cookie;
273 array->eof_index = array->size;
275 nfs_readdir_release_array(page);
280 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
282 loff_t diff = desc->file->f_pos - desc->current_index;
287 if (diff >= array->size) {
288 if (array->eof_index >= 0)
290 desc->current_index += array->size;
294 index = (unsigned int)diff;
295 *desc->dir_cookie = array->array[index].cookie;
296 desc->cache_entry_index = index;
304 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
307 int status = -EAGAIN;
309 for (i = 0; i < array->size; i++) {
310 if (array->array[i].cookie == *desc->dir_cookie) {
311 desc->cache_entry_index = i;
316 if (i == array->eof_index) {
318 status = -EBADCOOKIE;
325 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
327 struct nfs_cache_array *array;
328 int status = -EBADCOOKIE;
330 if (desc->dir_cookie == NULL)
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 nfs_readdir_release_array(desc->page);
349 /* Fill a page with xdr information before transferring to the cache page */
351 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
352 struct nfs_entry *entry, struct file *file, struct inode *inode)
354 struct rpc_cred *cred = nfs_file_cred(file);
355 unsigned long timestamp, gencount;
360 gencount = nfs_inc_attr_generation_counter();
361 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
362 NFS_SERVER(inode)->dtsize, desc->plus);
364 /* We requested READDIRPLUS, but the server doesn't grok it */
365 if (error == -ENOTSUPP && desc->plus) {
366 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
367 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
373 desc->timestamp = timestamp;
374 desc->gencount = gencount;
379 /* Fill in an entry based on the xdr code stored in desc->page */
381 int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream)
383 __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus);
387 entry->fattr->time_start = desc->timestamp;
388 entry->fattr->gencount = desc->gencount;
393 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
395 struct nfs_inode *node;
396 if (dentry->d_inode == NULL)
398 node = NFS_I(dentry->d_inode);
399 if (node->fh.size != entry->fh->size)
401 if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 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 == 1)) {
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->dir_cookie;
569 entry.fh = nfs_alloc_fhandle();
570 entry.fattr = nfs_alloc_fattr();
571 if (entry.fh == NULL || entry.fattr == NULL)
574 array = nfs_readdir_get_array(page);
576 status = PTR_ERR(array);
579 memset(array, 0, sizeof(struct nfs_cache_array));
580 array->eof_index = -1;
582 pages_ptr = nfs_readdir_large_page(pages, array_size);
584 goto out_release_array;
587 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
592 status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen);
594 if (status == -ENOSPC)
598 } while (array->eof_index < 0);
600 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
602 nfs_readdir_release_array(page);
604 nfs_free_fattr(entry.fattr);
605 nfs_free_fhandle(entry.fh);
610 * Now we cache directories properly, by converting xdr information
611 * to an array that can be used for lookups later. This results in
612 * fewer cache pages, since we can store more information on each page.
613 * We only need to convert from xdr once so future lookups are much simpler
616 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
618 struct inode *inode = desc->file->f_path.dentry->d_inode;
621 ret = nfs_readdir_xdr_to_array(desc, page, inode);
624 SetPageUptodate(page);
626 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
627 /* Should never happen */
628 nfs_zap_mapping(inode, inode->i_mapping);
638 void cache_page_release(nfs_readdir_descriptor_t *desc)
640 page_cache_release(desc->page);
645 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
647 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
648 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
652 * Returns 0 if desc->dir_cookie was found on page desc->page_index
655 int find_cache_page(nfs_readdir_descriptor_t *desc)
659 desc->page = get_cache_page(desc);
660 if (IS_ERR(desc->page))
661 return PTR_ERR(desc->page);
663 res = nfs_readdir_search_array(desc);
666 cache_page_release(desc);
670 /* Search for desc->dir_cookie from the beginning of the page cache */
672 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
676 if (desc->page_index == 0)
677 desc->current_index = 0;
679 res = find_cache_page(desc);
687 static inline unsigned int dt_type(struct inode *inode)
689 return (inode->i_mode >> 12) & 15;
693 * Once we've found the start of the dirent within a page: fill 'er up...
696 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
699 struct file *file = desc->file;
702 struct nfs_cache_array *array = NULL;
703 unsigned int d_type = DT_UNKNOWN;
705 array = nfs_readdir_get_array(desc->page);
707 res = PTR_ERR(array);
711 for (i = desc->cache_entry_index; i < array->size; i++) {
714 res = filldir(dirent, array->array[i].string.name,
715 array->array[i].string.len, file->f_pos,
716 nfs_compat_user_ino64(array->array[i].ino), d_type);
720 desc->cache_entry_index = i;
721 if (i < (array->size-1))
722 *desc->dir_cookie = array->array[i+1].cookie;
724 *desc->dir_cookie = array->last_cookie;
726 if (i == array->eof_index)
729 nfs_readdir_release_array(desc->page);
731 cache_page_release(desc);
732 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
733 (unsigned long long)*desc->dir_cookie, res);
738 * If we cannot find a cookie in our cache, we suspect that this is
739 * because it points to a deleted file, so we ask the server to return
740 * whatever it thinks is the next entry. We then feed this to filldir.
741 * If all goes well, we should then be able to find our way round the
742 * cache on the next call to readdir_search_pagecache();
744 * NOTE: we cannot add the anonymous page to the pagecache because
745 * the data it contains might not be page aligned. Besides,
746 * we should already have a complete representation of the
747 * directory in the page cache by the time we get here.
750 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
753 struct page *page = NULL;
755 struct inode *inode = desc->file->f_path.dentry->d_inode;
757 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
758 (unsigned long long)*desc->dir_cookie);
760 page = alloc_page(GFP_HIGHUSER);
766 desc->page_index = 0;
769 if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) {
774 status = nfs_do_filldir(desc, dirent, filldir);
777 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
781 cache_page_release(desc);
785 /* The file offset position represents the dirent entry number. A
786 last cookie cache takes care of the common case of reading the
789 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
791 struct dentry *dentry = filp->f_path.dentry;
792 struct inode *inode = dentry->d_inode;
793 nfs_readdir_descriptor_t my_desc,
797 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
798 dentry->d_parent->d_name.name, dentry->d_name.name,
799 (long long)filp->f_pos);
800 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
803 * filp->f_pos points to the dirent entry number.
804 * *desc->dir_cookie has the cookie for the next entry. We have
805 * to either find the entry with the appropriate number or
806 * revalidate the cookie.
808 memset(desc, 0, sizeof(*desc));
811 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
812 desc->decode = NFS_PROTO(inode)->decode_dirent;
813 desc->plus = NFS_USE_READDIRPLUS(inode);
815 nfs_block_sillyrename(dentry);
816 res = nfs_revalidate_mapping(inode, filp->f_mapping);
820 while (desc->eof != 1) {
821 res = readdir_search_pagecache(desc);
823 if (res == -EBADCOOKIE) {
824 /* This means either end of directory */
825 if (*desc->dir_cookie && desc->eof == 0) {
826 /* Or that the server has 'lost' a cookie */
827 res = uncached_readdir(desc, dirent, filldir);
834 if (res == -ETOOSMALL && desc->plus) {
835 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
836 nfs_zap_caches(inode);
837 desc->page_index = 0;
845 res = nfs_do_filldir(desc, dirent, filldir);
852 nfs_unblock_sillyrename(dentry);
855 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
856 dentry->d_parent->d_name.name, dentry->d_name.name,
861 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
863 struct dentry *dentry = filp->f_path.dentry;
864 struct inode *inode = dentry->d_inode;
866 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
867 dentry->d_parent->d_name.name,
871 mutex_lock(&inode->i_mutex);
874 offset += filp->f_pos;
882 if (offset != filp->f_pos) {
883 filp->f_pos = offset;
884 nfs_file_open_context(filp)->dir_cookie = 0;
887 mutex_unlock(&inode->i_mutex);
892 * All directory operations under NFS are synchronous, so fsync()
893 * is a dummy operation.
895 static int nfs_fsync_dir(struct file *filp, int datasync)
897 struct dentry *dentry = filp->f_path.dentry;
899 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
900 dentry->d_parent->d_name.name, dentry->d_name.name,
903 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
908 * nfs_force_lookup_revalidate - Mark the directory as having changed
909 * @dir - pointer to directory inode
911 * This forces the revalidation code in nfs_lookup_revalidate() to do a
912 * full lookup on all child dentries of 'dir' whenever a change occurs
913 * on the server that might have invalidated our dcache.
915 * The caller should be holding dir->i_lock
917 void nfs_force_lookup_revalidate(struct inode *dir)
919 NFS_I(dir)->cache_change_attribute++;
923 * A check for whether or not the parent directory has changed.
924 * In the case it has, we assume that the dentries are untrustworthy
925 * and may need to be looked up again.
927 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
931 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
933 if (!nfs_verify_change_attribute(dir, dentry->d_time))
935 /* Revalidate nfsi->cache_change_attribute before we declare a match */
936 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
938 if (!nfs_verify_change_attribute(dir, dentry->d_time))
944 * Return the intent data that applies to this particular path component
946 * Note that the current set of intents only apply to the very last
947 * component of the path.
948 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
950 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
952 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
954 return nd->flags & mask;
958 * Use intent information to check whether or not we're going to do
959 * an O_EXCL create using this path component.
961 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
963 if (NFS_PROTO(dir)->version == 2)
965 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
969 * Inode and filehandle revalidation for lookups.
971 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
972 * or if the intent information indicates that we're about to open this
973 * particular file and the "nocto" mount flag is not set.
977 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
979 struct nfs_server *server = NFS_SERVER(inode);
981 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
984 /* VFS wants an on-the-wire revalidation */
985 if (nd->flags & LOOKUP_REVAL)
987 /* This is an open(2) */
988 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
989 !(server->flags & NFS_MOUNT_NOCTO) &&
990 (S_ISREG(inode->i_mode) ||
991 S_ISDIR(inode->i_mode)))
995 return nfs_revalidate_inode(server, inode);
997 return __nfs_revalidate_inode(server, inode);
1001 * We judge how long we want to trust negative
1002 * dentries by looking at the parent inode mtime.
1004 * If parent mtime has changed, we revalidate, else we wait for a
1005 * period corresponding to the parent's attribute cache timeout value.
1008 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1009 struct nameidata *nd)
1011 /* Don't revalidate a negative dentry if we're creating a new file */
1012 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1014 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1016 return !nfs_check_verifier(dir, dentry);
1020 * This is called every time the dcache has a lookup hit,
1021 * and we should check whether we can really trust that
1024 * NOTE! The hit can be a negative hit too, don't assume
1027 * If the parent directory is seen to have changed, we throw out the
1028 * cached dentry and do a new lookup.
1030 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
1033 struct inode *inode;
1034 struct dentry *parent;
1035 struct nfs_fh *fhandle = NULL;
1036 struct nfs_fattr *fattr = NULL;
1039 parent = dget_parent(dentry);
1040 dir = parent->d_inode;
1041 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1042 inode = dentry->d_inode;
1045 if (nfs_neg_need_reval(dir, dentry, nd))
1050 if (is_bad_inode(inode)) {
1051 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1052 __func__, dentry->d_parent->d_name.name,
1053 dentry->d_name.name);
1057 if (nfs_have_delegation(inode, FMODE_READ))
1058 goto out_set_verifier;
1060 /* Force a full look up iff the parent directory has changed */
1061 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1062 if (nfs_lookup_verify_inode(inode, nd))
1063 goto out_zap_parent;
1067 if (NFS_STALE(inode))
1071 fhandle = nfs_alloc_fhandle();
1072 fattr = nfs_alloc_fattr();
1073 if (fhandle == NULL || fattr == NULL)
1076 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1079 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1081 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1084 nfs_free_fattr(fattr);
1085 nfs_free_fhandle(fhandle);
1087 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1090 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1091 __func__, dentry->d_parent->d_name.name,
1092 dentry->d_name.name);
1095 nfs_zap_caches(dir);
1097 nfs_mark_for_revalidate(dir);
1098 if (inode && S_ISDIR(inode->i_mode)) {
1099 /* Purge readdir caches. */
1100 nfs_zap_caches(inode);
1101 /* If we have submounts, don't unhash ! */
1102 if (have_submounts(dentry))
1104 if (dentry->d_flags & DCACHE_DISCONNECTED)
1106 shrink_dcache_parent(dentry);
1109 nfs_free_fattr(fattr);
1110 nfs_free_fhandle(fhandle);
1112 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1113 __func__, dentry->d_parent->d_name.name,
1114 dentry->d_name.name);
1117 nfs_free_fattr(fattr);
1118 nfs_free_fhandle(fhandle);
1120 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1121 __func__, dentry->d_parent->d_name.name,
1122 dentry->d_name.name, error);
1127 * This is called from dput() when d_count is going to 0.
1129 static int nfs_dentry_delete(struct dentry *dentry)
1131 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1132 dentry->d_parent->d_name.name, dentry->d_name.name,
1135 /* Unhash any dentry with a stale inode */
1136 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1139 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1140 /* Unhash it, so that ->d_iput() would be called */
1143 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1144 /* Unhash it, so that ancestors of killed async unlink
1145 * files will be cleaned up during umount */
1152 static void nfs_drop_nlink(struct inode *inode)
1154 spin_lock(&inode->i_lock);
1155 if (inode->i_nlink > 0)
1157 spin_unlock(&inode->i_lock);
1161 * Called when the dentry loses inode.
1162 * We use it to clean up silly-renamed files.
1164 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1166 if (S_ISDIR(inode->i_mode))
1167 /* drop any readdir cache as it could easily be old */
1168 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1170 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1172 nfs_complete_unlink(dentry, inode);
1177 const struct dentry_operations nfs_dentry_operations = {
1178 .d_revalidate = nfs_lookup_revalidate,
1179 .d_delete = nfs_dentry_delete,
1180 .d_iput = nfs_dentry_iput,
1183 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1186 struct dentry *parent;
1187 struct inode *inode = NULL;
1188 struct nfs_fh *fhandle = NULL;
1189 struct nfs_fattr *fattr = NULL;
1192 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1193 dentry->d_parent->d_name.name, dentry->d_name.name);
1194 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1196 res = ERR_PTR(-ENAMETOOLONG);
1197 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1200 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1203 * If we're doing an exclusive create, optimize away the lookup
1204 * but don't hash the dentry.
1206 if (nfs_is_exclusive_create(dir, nd)) {
1207 d_instantiate(dentry, NULL);
1212 res = ERR_PTR(-ENOMEM);
1213 fhandle = nfs_alloc_fhandle();
1214 fattr = nfs_alloc_fattr();
1215 if (fhandle == NULL || fattr == NULL)
1218 parent = dentry->d_parent;
1219 /* Protect against concurrent sillydeletes */
1220 nfs_block_sillyrename(parent);
1221 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1222 if (error == -ENOENT)
1225 res = ERR_PTR(error);
1226 goto out_unblock_sillyrename;
1228 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1229 res = (struct dentry *)inode;
1231 goto out_unblock_sillyrename;
1234 res = d_materialise_unique(dentry, inode);
1237 goto out_unblock_sillyrename;
1240 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1241 out_unblock_sillyrename:
1242 nfs_unblock_sillyrename(parent);
1244 nfs_free_fattr(fattr);
1245 nfs_free_fhandle(fhandle);
1249 #ifdef CONFIG_NFS_V4
1250 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1252 const struct dentry_operations nfs4_dentry_operations = {
1253 .d_revalidate = nfs_open_revalidate,
1254 .d_delete = nfs_dentry_delete,
1255 .d_iput = nfs_dentry_iput,
1259 * Use intent information to determine whether we need to substitute
1260 * the NFSv4-style stateful OPEN for the LOOKUP call
1262 static int is_atomic_open(struct nameidata *nd)
1264 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1266 /* NFS does not (yet) have a stateful open for directories */
1267 if (nd->flags & LOOKUP_DIRECTORY)
1269 /* Are we trying to write to a read only partition? */
1270 if (__mnt_is_readonly(nd->path.mnt) &&
1271 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1276 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1278 struct path path = {
1279 .mnt = nd->path.mnt,
1282 struct nfs_open_context *ctx;
1283 struct rpc_cred *cred;
1284 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1286 cred = rpc_lookup_cred();
1288 return ERR_CAST(cred);
1289 ctx = alloc_nfs_open_context(&path, cred, fmode);
1292 return ERR_PTR(-ENOMEM);
1296 static int do_open(struct inode *inode, struct file *filp)
1298 nfs_fscache_set_inode_cookie(inode, filp);
1302 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1307 /* If the open_intent is for execute, we have an extra check to make */
1308 if (ctx->mode & FMODE_EXEC) {
1309 ret = nfs_may_open(ctx->path.dentry->d_inode,
1311 nd->intent.open.flags);
1315 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1317 ret = PTR_ERR(filp);
1319 nfs_file_set_open_context(filp, ctx);
1321 put_nfs_open_context(ctx);
1325 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1327 struct nfs_open_context *ctx;
1329 struct dentry *res = NULL;
1330 struct inode *inode;
1334 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1335 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1337 /* Check that we are indeed trying to open this file */
1338 if (!is_atomic_open(nd))
1341 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1342 res = ERR_PTR(-ENAMETOOLONG);
1345 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1347 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1349 if (nd->flags & LOOKUP_EXCL) {
1350 d_instantiate(dentry, NULL);
1354 ctx = nameidata_to_nfs_open_context(dentry, nd);
1355 res = ERR_CAST(ctx);
1359 open_flags = nd->intent.open.flags;
1360 if (nd->flags & LOOKUP_CREATE) {
1361 attr.ia_mode = nd->intent.open.create_mode;
1362 attr.ia_valid = ATTR_MODE;
1363 if (!IS_POSIXACL(dir))
1364 attr.ia_mode &= ~current_umask();
1366 open_flags &= ~(O_EXCL | O_CREAT);
1370 /* Open the file on the server */
1371 nfs_block_sillyrename(dentry->d_parent);
1372 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1373 if (IS_ERR(inode)) {
1374 nfs_unblock_sillyrename(dentry->d_parent);
1375 put_nfs_open_context(ctx);
1376 switch (PTR_ERR(inode)) {
1377 /* Make a negative dentry */
1379 d_add(dentry, NULL);
1382 /* This turned out not to be a regular file */
1386 if (!(nd->intent.open.flags & O_NOFOLLOW))
1391 res = ERR_CAST(inode);
1395 res = d_add_unique(dentry, inode);
1396 nfs_unblock_sillyrename(dentry->d_parent);
1398 dput(ctx->path.dentry);
1399 ctx->path.dentry = dget(res);
1402 err = nfs_intent_set_file(nd, ctx);
1406 return ERR_PTR(err);
1409 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1412 return nfs_lookup(dir, dentry, nd);
1415 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1417 struct dentry *parent = NULL;
1418 struct inode *inode = dentry->d_inode;
1420 struct nfs_open_context *ctx;
1421 int openflags, ret = 0;
1423 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1426 parent = dget_parent(dentry);
1427 dir = parent->d_inode;
1429 /* We can't create new files in nfs_open_revalidate(), so we
1430 * optimize away revalidation of negative dentries.
1432 if (inode == NULL) {
1433 if (!nfs_neg_need_reval(dir, dentry, nd))
1438 /* NFS only supports OPEN on regular files */
1439 if (!S_ISREG(inode->i_mode))
1441 openflags = nd->intent.open.flags;
1442 /* We cannot do exclusive creation on a positive dentry */
1443 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1445 /* We can't create new files, or truncate existing ones here */
1446 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1448 ctx = nameidata_to_nfs_open_context(dentry, nd);
1453 * Note: we're not holding inode->i_mutex and so may be racing with
1454 * operations that change the directory. We therefore save the
1455 * change attribute *before* we do the RPC call.
1457 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1458 if (IS_ERR(inode)) {
1459 ret = PTR_ERR(inode);
1472 if (inode != dentry->d_inode)
1475 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1476 ret = nfs_intent_set_file(nd, ctx);
1486 put_nfs_open_context(ctx);
1492 return nfs_lookup_revalidate(dentry, nd);
1495 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1496 struct nameidata *nd)
1498 struct nfs_open_context *ctx = NULL;
1503 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1504 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1506 attr.ia_mode = mode;
1507 attr.ia_valid = ATTR_MODE;
1509 if ((nd->flags & LOOKUP_CREATE) != 0) {
1510 open_flags = nd->intent.open.flags;
1512 ctx = nameidata_to_nfs_open_context(dentry, nd);
1513 error = PTR_ERR(ctx);
1518 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1522 error = nfs_intent_set_file(nd, ctx);
1529 put_nfs_open_context(ctx);
1536 #endif /* CONFIG_NFSV4 */
1539 * Code common to create, mkdir, and mknod.
1541 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1542 struct nfs_fattr *fattr)
1544 struct dentry *parent = dget_parent(dentry);
1545 struct inode *dir = parent->d_inode;
1546 struct inode *inode;
1547 int error = -EACCES;
1551 /* We may have been initialized further down */
1552 if (dentry->d_inode)
1554 if (fhandle->size == 0) {
1555 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1559 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1560 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1561 struct nfs_server *server = NFS_SB(dentry->d_sb);
1562 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1566 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1567 error = PTR_ERR(inode);
1570 d_add(dentry, inode);
1575 nfs_mark_for_revalidate(dir);
1581 * Following a failed create operation, we drop the dentry rather
1582 * than retain a negative dentry. This avoids a problem in the event
1583 * that the operation succeeded on the server, but an error in the
1584 * reply path made it appear to have failed.
1586 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1587 struct nameidata *nd)
1592 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1593 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1595 attr.ia_mode = mode;
1596 attr.ia_valid = ATTR_MODE;
1598 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1608 * See comments for nfs_proc_create regarding failed operations.
1611 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1616 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1617 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1619 if (!new_valid_dev(rdev))
1622 attr.ia_mode = mode;
1623 attr.ia_valid = ATTR_MODE;
1625 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1635 * See comments for nfs_proc_create regarding failed operations.
1637 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1642 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1643 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1645 attr.ia_valid = ATTR_MODE;
1646 attr.ia_mode = mode | S_IFDIR;
1648 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1657 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1659 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1663 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1667 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1668 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1670 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1671 /* Ensure the VFS deletes this inode */
1672 if (error == 0 && dentry->d_inode != NULL)
1673 clear_nlink(dentry->d_inode);
1674 else if (error == -ENOENT)
1675 nfs_dentry_handle_enoent(dentry);
1681 * Remove a file after making sure there are no pending writes,
1682 * and after checking that the file has only one user.
1684 * We invalidate the attribute cache and free the inode prior to the operation
1685 * to avoid possible races if the server reuses the inode.
1687 static int nfs_safe_remove(struct dentry *dentry)
1689 struct inode *dir = dentry->d_parent->d_inode;
1690 struct inode *inode = dentry->d_inode;
1693 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1694 dentry->d_parent->d_name.name, dentry->d_name.name);
1696 /* If the dentry was sillyrenamed, we simply call d_delete() */
1697 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1702 if (inode != NULL) {
1703 nfs_inode_return_delegation(inode);
1704 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1705 /* The VFS may want to delete this inode */
1707 nfs_drop_nlink(inode);
1708 nfs_mark_for_revalidate(inode);
1710 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1711 if (error == -ENOENT)
1712 nfs_dentry_handle_enoent(dentry);
1717 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1718 * belongs to an active ".nfs..." file and we return -EBUSY.
1720 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1722 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1725 int need_rehash = 0;
1727 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1728 dir->i_ino, dentry->d_name.name);
1730 spin_lock(&dcache_lock);
1731 spin_lock(&dentry->d_lock);
1732 if (atomic_read(&dentry->d_count) > 1) {
1733 spin_unlock(&dentry->d_lock);
1734 spin_unlock(&dcache_lock);
1735 /* Start asynchronous writeout of the inode */
1736 write_inode_now(dentry->d_inode, 0);
1737 error = nfs_sillyrename(dir, dentry);
1740 if (!d_unhashed(dentry)) {
1744 spin_unlock(&dentry->d_lock);
1745 spin_unlock(&dcache_lock);
1746 error = nfs_safe_remove(dentry);
1747 if (!error || error == -ENOENT) {
1748 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1749 } else if (need_rehash)
1755 * To create a symbolic link, most file systems instantiate a new inode,
1756 * add a page to it containing the path, then write it out to the disk
1757 * using prepare_write/commit_write.
1759 * Unfortunately the NFS client can't create the in-core inode first
1760 * because it needs a file handle to create an in-core inode (see
1761 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1762 * symlink request has completed on the server.
1764 * So instead we allocate a raw page, copy the symname into it, then do
1765 * the SYMLINK request with the page as the buffer. If it succeeds, we
1766 * now have a new file handle and can instantiate an in-core NFS inode
1767 * and move the raw page into its mapping.
1769 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1771 struct pagevec lru_pvec;
1775 unsigned int pathlen = strlen(symname);
1778 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1779 dir->i_ino, dentry->d_name.name, symname);
1781 if (pathlen > PAGE_SIZE)
1782 return -ENAMETOOLONG;
1784 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1785 attr.ia_valid = ATTR_MODE;
1787 page = alloc_page(GFP_HIGHUSER);
1791 kaddr = kmap_atomic(page, KM_USER0);
1792 memcpy(kaddr, symname, pathlen);
1793 if (pathlen < PAGE_SIZE)
1794 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1795 kunmap_atomic(kaddr, KM_USER0);
1797 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1799 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1800 dir->i_sb->s_id, dir->i_ino,
1801 dentry->d_name.name, symname, error);
1808 * No big deal if we can't add this page to the page cache here.
1809 * READLINK will get the missing page from the server if needed.
1811 pagevec_init(&lru_pvec, 0);
1812 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1814 pagevec_add(&lru_pvec, page);
1815 pagevec_lru_add_file(&lru_pvec);
1816 SetPageUptodate(page);
1825 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1827 struct inode *inode = old_dentry->d_inode;
1830 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1831 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1832 dentry->d_parent->d_name.name, dentry->d_name.name);
1834 nfs_inode_return_delegation(inode);
1837 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1840 d_add(dentry, inode);
1847 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1848 * different file handle for the same inode after a rename (e.g. when
1849 * moving to a different directory). A fail-safe method to do so would
1850 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1851 * rename the old file using the sillyrename stuff. This way, the original
1852 * file in old_dir will go away when the last process iput()s the inode.
1856 * It actually works quite well. One needs to have the possibility for
1857 * at least one ".nfs..." file in each directory the file ever gets
1858 * moved or linked to which happens automagically with the new
1859 * implementation that only depends on the dcache stuff instead of
1860 * using the inode layer
1862 * Unfortunately, things are a little more complicated than indicated
1863 * above. For a cross-directory move, we want to make sure we can get
1864 * rid of the old inode after the operation. This means there must be
1865 * no pending writes (if it's a file), and the use count must be 1.
1866 * If these conditions are met, we can drop the dentries before doing
1869 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1870 struct inode *new_dir, struct dentry *new_dentry)
1872 struct inode *old_inode = old_dentry->d_inode;
1873 struct inode *new_inode = new_dentry->d_inode;
1874 struct dentry *dentry = NULL, *rehash = NULL;
1877 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1878 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1879 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1880 atomic_read(&new_dentry->d_count));
1883 * For non-directories, check whether the target is busy and if so,
1884 * make a copy of the dentry and then do a silly-rename. If the
1885 * silly-rename succeeds, the copied dentry is hashed and becomes
1888 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1890 * To prevent any new references to the target during the
1891 * rename, we unhash the dentry in advance.
1893 if (!d_unhashed(new_dentry)) {
1895 rehash = new_dentry;
1898 if (atomic_read(&new_dentry->d_count) > 2) {
1901 /* copy the target dentry's name */
1902 dentry = d_alloc(new_dentry->d_parent,
1903 &new_dentry->d_name);
1907 /* silly-rename the existing target ... */
1908 err = nfs_sillyrename(new_dir, new_dentry);
1912 new_dentry = dentry;
1918 nfs_inode_return_delegation(old_inode);
1919 if (new_inode != NULL)
1920 nfs_inode_return_delegation(new_inode);
1922 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1923 new_dir, &new_dentry->d_name);
1924 nfs_mark_for_revalidate(old_inode);
1929 if (new_inode != NULL)
1930 nfs_drop_nlink(new_inode);
1931 d_move(old_dentry, new_dentry);
1932 nfs_set_verifier(new_dentry,
1933 nfs_save_change_attribute(new_dir));
1934 } else if (error == -ENOENT)
1935 nfs_dentry_handle_enoent(old_dentry);
1937 /* new dentry created? */
1943 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1944 static LIST_HEAD(nfs_access_lru_list);
1945 static atomic_long_t nfs_access_nr_entries;
1947 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1949 put_rpccred(entry->cred);
1951 smp_mb__before_atomic_dec();
1952 atomic_long_dec(&nfs_access_nr_entries);
1953 smp_mb__after_atomic_dec();
1956 static void nfs_access_free_list(struct list_head *head)
1958 struct nfs_access_entry *cache;
1960 while (!list_empty(head)) {
1961 cache = list_entry(head->next, struct nfs_access_entry, lru);
1962 list_del(&cache->lru);
1963 nfs_access_free_entry(cache);
1967 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1970 struct nfs_inode *nfsi, *next;
1971 struct nfs_access_entry *cache;
1973 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1974 return (nr_to_scan == 0) ? 0 : -1;
1976 spin_lock(&nfs_access_lru_lock);
1977 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1978 struct inode *inode;
1980 if (nr_to_scan-- == 0)
1982 inode = &nfsi->vfs_inode;
1983 spin_lock(&inode->i_lock);
1984 if (list_empty(&nfsi->access_cache_entry_lru))
1985 goto remove_lru_entry;
1986 cache = list_entry(nfsi->access_cache_entry_lru.next,
1987 struct nfs_access_entry, lru);
1988 list_move(&cache->lru, &head);
1989 rb_erase(&cache->rb_node, &nfsi->access_cache);
1990 if (!list_empty(&nfsi->access_cache_entry_lru))
1991 list_move_tail(&nfsi->access_cache_inode_lru,
1992 &nfs_access_lru_list);
1995 list_del_init(&nfsi->access_cache_inode_lru);
1996 smp_mb__before_clear_bit();
1997 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1998 smp_mb__after_clear_bit();
2000 spin_unlock(&inode->i_lock);
2002 spin_unlock(&nfs_access_lru_lock);
2003 nfs_access_free_list(&head);
2004 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2007 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2009 struct rb_root *root_node = &nfsi->access_cache;
2011 struct nfs_access_entry *entry;
2013 /* Unhook entries from the cache */
2014 while ((n = rb_first(root_node)) != NULL) {
2015 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2016 rb_erase(n, root_node);
2017 list_move(&entry->lru, head);
2019 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2022 void nfs_access_zap_cache(struct inode *inode)
2026 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2028 /* Remove from global LRU init */
2029 spin_lock(&nfs_access_lru_lock);
2030 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2031 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2033 spin_lock(&inode->i_lock);
2034 __nfs_access_zap_cache(NFS_I(inode), &head);
2035 spin_unlock(&inode->i_lock);
2036 spin_unlock(&nfs_access_lru_lock);
2037 nfs_access_free_list(&head);
2040 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2042 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2043 struct nfs_access_entry *entry;
2046 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2048 if (cred < entry->cred)
2050 else if (cred > entry->cred)
2058 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2060 struct nfs_inode *nfsi = NFS_I(inode);
2061 struct nfs_access_entry *cache;
2064 spin_lock(&inode->i_lock);
2065 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2067 cache = nfs_access_search_rbtree(inode, cred);
2070 if (!nfs_have_delegated_attributes(inode) &&
2071 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2073 res->jiffies = cache->jiffies;
2074 res->cred = cache->cred;
2075 res->mask = cache->mask;
2076 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2079 spin_unlock(&inode->i_lock);
2082 rb_erase(&cache->rb_node, &nfsi->access_cache);
2083 list_del(&cache->lru);
2084 spin_unlock(&inode->i_lock);
2085 nfs_access_free_entry(cache);
2088 spin_unlock(&inode->i_lock);
2089 nfs_access_zap_cache(inode);
2093 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2095 struct nfs_inode *nfsi = NFS_I(inode);
2096 struct rb_root *root_node = &nfsi->access_cache;
2097 struct rb_node **p = &root_node->rb_node;
2098 struct rb_node *parent = NULL;
2099 struct nfs_access_entry *entry;
2101 spin_lock(&inode->i_lock);
2102 while (*p != NULL) {
2104 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2106 if (set->cred < entry->cred)
2107 p = &parent->rb_left;
2108 else if (set->cred > entry->cred)
2109 p = &parent->rb_right;
2113 rb_link_node(&set->rb_node, parent, p);
2114 rb_insert_color(&set->rb_node, root_node);
2115 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2116 spin_unlock(&inode->i_lock);
2119 rb_replace_node(parent, &set->rb_node, root_node);
2120 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2121 list_del(&entry->lru);
2122 spin_unlock(&inode->i_lock);
2123 nfs_access_free_entry(entry);
2126 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2128 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2131 RB_CLEAR_NODE(&cache->rb_node);
2132 cache->jiffies = set->jiffies;
2133 cache->cred = get_rpccred(set->cred);
2134 cache->mask = set->mask;
2136 nfs_access_add_rbtree(inode, cache);
2138 /* Update accounting */
2139 smp_mb__before_atomic_inc();
2140 atomic_long_inc(&nfs_access_nr_entries);
2141 smp_mb__after_atomic_inc();
2143 /* Add inode to global LRU list */
2144 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2145 spin_lock(&nfs_access_lru_lock);
2146 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2147 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2148 &nfs_access_lru_list);
2149 spin_unlock(&nfs_access_lru_lock);
2153 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2155 struct nfs_access_entry cache;
2158 status = nfs_access_get_cached(inode, cred, &cache);
2162 /* Be clever: ask server to check for all possible rights */
2163 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2165 cache.jiffies = jiffies;
2166 status = NFS_PROTO(inode)->access(inode, &cache);
2168 if (status == -ESTALE) {
2169 nfs_zap_caches(inode);
2170 if (!S_ISDIR(inode->i_mode))
2171 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2175 nfs_access_add_cache(inode, &cache);
2177 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2182 static int nfs_open_permission_mask(int openflags)
2186 if (openflags & FMODE_READ)
2188 if (openflags & FMODE_WRITE)
2190 if (openflags & FMODE_EXEC)
2195 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2197 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2200 int nfs_permission(struct inode *inode, int mask)
2202 struct rpc_cred *cred;
2205 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2207 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2209 /* Is this sys_access() ? */
2210 if (mask & (MAY_ACCESS | MAY_CHDIR))
2213 switch (inode->i_mode & S_IFMT) {
2217 /* NFSv4 has atomic_open... */
2218 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2219 && (mask & MAY_OPEN)
2220 && !(mask & MAY_EXEC))
2225 * Optimize away all write operations, since the server
2226 * will check permissions when we perform the op.
2228 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2233 if (!NFS_PROTO(inode)->access)
2236 cred = rpc_lookup_cred();
2237 if (!IS_ERR(cred)) {
2238 res = nfs_do_access(inode, cred, mask);
2241 res = PTR_ERR(cred);
2243 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2246 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2247 inode->i_sb->s_id, inode->i_ino, mask, res);
2250 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2252 res = generic_permission(inode, mask, NULL);
2258 * version-control: t
2259 * kept-new-versions: 5