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 void nfs_readdir_clear_array(struct page*);
62 const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .readdir = nfs_readdir,
67 .release = nfs_release,
68 .fsync = nfs_fsync_dir,
71 const struct inode_operations nfs_dir_inode_operations = {
76 .symlink = nfs_symlink,
81 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
86 const struct address_space_operations nfs_dir_aops = {
87 .freepage = nfs_readdir_clear_array,
91 const struct inode_operations nfs3_dir_inode_operations = {
96 .symlink = nfs_symlink,
100 .rename = nfs_rename,
101 .permission = nfs_permission,
102 .getattr = nfs_getattr,
103 .setattr = nfs_setattr,
104 .listxattr = nfs3_listxattr,
105 .getxattr = nfs3_getxattr,
106 .setxattr = nfs3_setxattr,
107 .removexattr = nfs3_removexattr,
109 #endif /* CONFIG_NFS_V3 */
113 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
114 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
115 const struct inode_operations nfs4_dir_inode_operations = {
116 .create = nfs_open_create,
117 .lookup = nfs_atomic_lookup,
119 .unlink = nfs_unlink,
120 .symlink = nfs_symlink,
124 .rename = nfs_rename,
125 .permission = nfs_permission,
126 .getattr = nfs_getattr,
127 .setattr = nfs_setattr,
128 .getxattr = 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 {
165 unsigned char d_type;
168 struct nfs_cache_array {
172 struct nfs_cache_array_entry array[0];
175 typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int);
179 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 void nfs_readdir_clear_array(struct page *page)
219 struct nfs_cache_array *array;
222 array = kmap_atomic(page, KM_USER0);
223 for (i = 0; i < array->size; i++)
224 kfree(array->array[i].string.name);
225 kunmap_atomic(array, KM_USER0);
229 * the caller is responsible for freeing qstr.name
230 * when called by nfs_readdir_add_to_array, the strings will be freed in
231 * nfs_clear_readdir_array()
234 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
237 string->name = kmemdup(name, len, GFP_KERNEL);
238 if (string->name == NULL)
241 * Avoid a kmemleak false positive. The pointer to the name is stored
242 * in a page cache page which kmemleak does not scan.
244 kmemleak_not_leak(string->name);
245 string->hash = full_name_hash(name, len);
250 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
252 struct nfs_cache_array *array = nfs_readdir_get_array(page);
253 struct nfs_cache_array_entry *cache_entry;
257 return PTR_ERR(array);
259 cache_entry = &array->array[array->size];
261 /* Check that this entry lies within the page bounds */
263 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
266 cache_entry->cookie = entry->prev_cookie;
267 cache_entry->ino = entry->ino;
268 cache_entry->d_type = entry->d_type;
269 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
272 array->last_cookie = entry->cookie;
275 array->eof_index = array->size;
277 nfs_readdir_release_array(page);
282 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
284 loff_t diff = desc->file->f_pos - desc->current_index;
289 if (diff >= array->size) {
290 if (array->eof_index >= 0)
292 desc->current_index += array->size;
296 index = (unsigned int)diff;
297 *desc->dir_cookie = array->array[index].cookie;
298 desc->cache_entry_index = index;
306 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
309 int status = -EAGAIN;
311 for (i = 0; i < array->size; i++) {
312 if (array->array[i].cookie == *desc->dir_cookie) {
313 desc->cache_entry_index = i;
318 if (i == array->eof_index) {
320 status = -EBADCOOKIE;
327 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
329 struct nfs_cache_array *array;
330 int status = -EBADCOOKIE;
332 if (desc->dir_cookie == NULL)
335 array = nfs_readdir_get_array(desc->page);
337 status = PTR_ERR(array);
341 if (*desc->dir_cookie == 0)
342 status = nfs_readdir_search_for_pos(array, desc);
344 status = nfs_readdir_search_for_cookie(array, desc);
346 if (status == -EAGAIN)
347 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 /* Fill in an entry based on the xdr code stored in desc->page */
385 int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream)
387 __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), 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 == 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->last_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 if (!desc->page->mapping)
641 nfs_readdir_clear_array(desc->page);
642 page_cache_release(desc->page);
647 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
649 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
650 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
654 * Returns 0 if desc->dir_cookie was found on page desc->page_index
657 int find_cache_page(nfs_readdir_descriptor_t *desc)
661 desc->page = get_cache_page(desc);
662 if (IS_ERR(desc->page))
663 return PTR_ERR(desc->page);
665 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);
691 static inline unsigned int dt_type(struct inode *inode)
693 return (inode->i_mode >> 12) & 15;
697 * Once we've found the start of the dirent within a page: fill 'er up...
700 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
703 struct file *file = desc->file;
706 struct nfs_cache_array *array = NULL;
708 array = nfs_readdir_get_array(desc->page);
710 res = PTR_ERR(array);
714 for (i = desc->cache_entry_index; i < array->size; i++) {
715 struct nfs_cache_array_entry *ent;
717 ent = &array->array[i];
718 if (filldir(dirent, ent->string.name, ent->string.len,
719 file->f_pos, nfs_compat_user_ino64(ent->ino),
725 desc->cache_entry_index = i;
726 if (i < (array->size-1))
727 *desc->dir_cookie = array->array[i+1].cookie;
729 *desc->dir_cookie = array->last_cookie;
731 if (i == array->eof_index)
734 nfs_readdir_release_array(desc->page);
736 cache_page_release(desc);
737 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
738 (unsigned long long)*desc->dir_cookie, res);
743 * If we cannot find a cookie in our cache, we suspect that this is
744 * because it points to a deleted file, so we ask the server to return
745 * whatever it thinks is the next entry. We then feed this to filldir.
746 * If all goes well, we should then be able to find our way round the
747 * cache on the next call to readdir_search_pagecache();
749 * NOTE: we cannot add the anonymous page to the pagecache because
750 * the data it contains might not be page aligned. Besides,
751 * we should already have a complete representation of the
752 * directory in the page cache by the time we get here.
755 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
758 struct page *page = NULL;
760 struct inode *inode = desc->file->f_path.dentry->d_inode;
762 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
763 (unsigned long long)*desc->dir_cookie);
765 page = alloc_page(GFP_HIGHUSER);
771 desc->page_index = 0;
772 desc->last_cookie = *desc->dir_cookie;
775 status = nfs_readdir_xdr_to_array(desc, page, inode);
779 status = nfs_do_filldir(desc, dirent, filldir);
782 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
786 cache_page_release(desc);
790 /* The file offset position represents the dirent entry number. A
791 last cookie cache takes care of the common case of reading the
794 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
796 struct dentry *dentry = filp->f_path.dentry;
797 struct inode *inode = dentry->d_inode;
798 nfs_readdir_descriptor_t my_desc,
802 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
803 dentry->d_parent->d_name.name, dentry->d_name.name,
804 (long long)filp->f_pos);
805 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
808 * filp->f_pos points to the dirent entry number.
809 * *desc->dir_cookie has the cookie for the next entry. We have
810 * to either find the entry with the appropriate number or
811 * revalidate the cookie.
813 memset(desc, 0, sizeof(*desc));
816 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
817 desc->decode = NFS_PROTO(inode)->decode_dirent;
818 desc->plus = NFS_USE_READDIRPLUS(inode);
820 nfs_block_sillyrename(dentry);
821 res = nfs_revalidate_mapping(inode, filp->f_mapping);
825 while (desc->eof != 1) {
826 res = readdir_search_pagecache(desc);
828 if (res == -EBADCOOKIE) {
830 /* This means either end of directory */
831 if (*desc->dir_cookie && desc->eof == 0) {
832 /* Or that the server has 'lost' a cookie */
833 res = uncached_readdir(desc, dirent, filldir);
839 if (res == -ETOOSMALL && desc->plus) {
840 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
841 nfs_zap_caches(inode);
842 desc->page_index = 0;
850 res = nfs_do_filldir(desc, dirent, filldir);
855 nfs_unblock_sillyrename(dentry);
858 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
859 dentry->d_parent->d_name.name, dentry->d_name.name,
864 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
866 struct dentry *dentry = filp->f_path.dentry;
867 struct inode *inode = dentry->d_inode;
869 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
870 dentry->d_parent->d_name.name,
874 mutex_lock(&inode->i_mutex);
877 offset += filp->f_pos;
885 if (offset != filp->f_pos) {
886 filp->f_pos = offset;
887 nfs_file_open_context(filp)->dir_cookie = 0;
890 mutex_unlock(&inode->i_mutex);
895 * All directory operations under NFS are synchronous, so fsync()
896 * is a dummy operation.
898 static int nfs_fsync_dir(struct file *filp, int datasync)
900 struct dentry *dentry = filp->f_path.dentry;
902 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
903 dentry->d_parent->d_name.name, dentry->d_name.name,
906 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
911 * nfs_force_lookup_revalidate - Mark the directory as having changed
912 * @dir - pointer to directory inode
914 * This forces the revalidation code in nfs_lookup_revalidate() to do a
915 * full lookup on all child dentries of 'dir' whenever a change occurs
916 * on the server that might have invalidated our dcache.
918 * The caller should be holding dir->i_lock
920 void nfs_force_lookup_revalidate(struct inode *dir)
922 NFS_I(dir)->cache_change_attribute++;
926 * A check for whether or not the parent directory has changed.
927 * In the case it has, we assume that the dentries are untrustworthy
928 * and may need to be looked up again.
930 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
934 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
936 if (!nfs_verify_change_attribute(dir, dentry->d_time))
938 /* Revalidate nfsi->cache_change_attribute before we declare a match */
939 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
941 if (!nfs_verify_change_attribute(dir, dentry->d_time))
947 * Return the intent data that applies to this particular path component
949 * Note that the current set of intents only apply to the very last
950 * component of the path.
951 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
953 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
955 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
957 return nd->flags & mask;
961 * Use intent information to check whether or not we're going to do
962 * an O_EXCL create using this path component.
964 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
966 if (NFS_PROTO(dir)->version == 2)
968 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
972 * Inode and filehandle revalidation for lookups.
974 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
975 * or if the intent information indicates that we're about to open this
976 * particular file and the "nocto" mount flag is not set.
980 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
982 struct nfs_server *server = NFS_SERVER(inode);
984 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
987 /* VFS wants an on-the-wire revalidation */
988 if (nd->flags & LOOKUP_REVAL)
990 /* This is an open(2) */
991 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
992 !(server->flags & NFS_MOUNT_NOCTO) &&
993 (S_ISREG(inode->i_mode) ||
994 S_ISDIR(inode->i_mode)))
998 return nfs_revalidate_inode(server, inode);
1000 return __nfs_revalidate_inode(server, inode);
1004 * We judge how long we want to trust negative
1005 * dentries by looking at the parent inode mtime.
1007 * If parent mtime has changed, we revalidate, else we wait for a
1008 * period corresponding to the parent's attribute cache timeout value.
1011 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1012 struct nameidata *nd)
1014 /* Don't revalidate a negative dentry if we're creating a new file */
1015 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1017 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1019 return !nfs_check_verifier(dir, dentry);
1023 * This is called every time the dcache has a lookup hit,
1024 * and we should check whether we can really trust that
1027 * NOTE! The hit can be a negative hit too, don't assume
1030 * If the parent directory is seen to have changed, we throw out the
1031 * cached dentry and do a new lookup.
1033 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
1036 struct inode *inode;
1037 struct dentry *parent;
1038 struct nfs_fh *fhandle = NULL;
1039 struct nfs_fattr *fattr = NULL;
1042 parent = dget_parent(dentry);
1043 dir = parent->d_inode;
1044 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1045 inode = dentry->d_inode;
1048 if (nfs_neg_need_reval(dir, dentry, nd))
1053 if (is_bad_inode(inode)) {
1054 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1055 __func__, dentry->d_parent->d_name.name,
1056 dentry->d_name.name);
1060 if (nfs_have_delegation(inode, FMODE_READ))
1061 goto out_set_verifier;
1063 /* Force a full look up iff the parent directory has changed */
1064 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1065 if (nfs_lookup_verify_inode(inode, nd))
1066 goto out_zap_parent;
1070 if (NFS_STALE(inode))
1074 fhandle = nfs_alloc_fhandle();
1075 fattr = nfs_alloc_fattr();
1076 if (fhandle == NULL || fattr == NULL)
1079 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1082 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1084 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1087 nfs_free_fattr(fattr);
1088 nfs_free_fhandle(fhandle);
1090 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1093 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1094 __func__, dentry->d_parent->d_name.name,
1095 dentry->d_name.name);
1098 nfs_zap_caches(dir);
1100 nfs_mark_for_revalidate(dir);
1101 if (inode && S_ISDIR(inode->i_mode)) {
1102 /* Purge readdir caches. */
1103 nfs_zap_caches(inode);
1104 /* If we have submounts, don't unhash ! */
1105 if (have_submounts(dentry))
1107 if (dentry->d_flags & DCACHE_DISCONNECTED)
1109 shrink_dcache_parent(dentry);
1112 nfs_free_fattr(fattr);
1113 nfs_free_fhandle(fhandle);
1115 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1116 __func__, dentry->d_parent->d_name.name,
1117 dentry->d_name.name);
1120 nfs_free_fattr(fattr);
1121 nfs_free_fhandle(fhandle);
1123 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1124 __func__, dentry->d_parent->d_name.name,
1125 dentry->d_name.name, error);
1130 * This is called from dput() when d_count is going to 0.
1132 static int nfs_dentry_delete(struct dentry *dentry)
1134 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1135 dentry->d_parent->d_name.name, dentry->d_name.name,
1138 /* Unhash any dentry with a stale inode */
1139 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1142 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1143 /* Unhash it, so that ->d_iput() would be called */
1146 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1147 /* Unhash it, so that ancestors of killed async unlink
1148 * files will be cleaned up during umount */
1155 static void nfs_drop_nlink(struct inode *inode)
1157 spin_lock(&inode->i_lock);
1158 if (inode->i_nlink > 0)
1160 spin_unlock(&inode->i_lock);
1164 * Called when the dentry loses inode.
1165 * We use it to clean up silly-renamed files.
1167 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1169 if (S_ISDIR(inode->i_mode))
1170 /* drop any readdir cache as it could easily be old */
1171 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1173 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1175 nfs_complete_unlink(dentry, inode);
1180 const struct dentry_operations nfs_dentry_operations = {
1181 .d_revalidate = nfs_lookup_revalidate,
1182 .d_delete = nfs_dentry_delete,
1183 .d_iput = nfs_dentry_iput,
1186 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1189 struct dentry *parent;
1190 struct inode *inode = NULL;
1191 struct nfs_fh *fhandle = NULL;
1192 struct nfs_fattr *fattr = NULL;
1195 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1196 dentry->d_parent->d_name.name, dentry->d_name.name);
1197 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1199 res = ERR_PTR(-ENAMETOOLONG);
1200 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1203 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1206 * If we're doing an exclusive create, optimize away the lookup
1207 * but don't hash the dentry.
1209 if (nfs_is_exclusive_create(dir, nd)) {
1210 d_instantiate(dentry, NULL);
1215 res = ERR_PTR(-ENOMEM);
1216 fhandle = nfs_alloc_fhandle();
1217 fattr = nfs_alloc_fattr();
1218 if (fhandle == NULL || fattr == NULL)
1221 parent = dentry->d_parent;
1222 /* Protect against concurrent sillydeletes */
1223 nfs_block_sillyrename(parent);
1224 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1225 if (error == -ENOENT)
1228 res = ERR_PTR(error);
1229 goto out_unblock_sillyrename;
1231 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1232 res = (struct dentry *)inode;
1234 goto out_unblock_sillyrename;
1237 res = d_materialise_unique(dentry, inode);
1240 goto out_unblock_sillyrename;
1243 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1244 out_unblock_sillyrename:
1245 nfs_unblock_sillyrename(parent);
1247 nfs_free_fattr(fattr);
1248 nfs_free_fhandle(fhandle);
1252 #ifdef CONFIG_NFS_V4
1253 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1255 const struct dentry_operations nfs4_dentry_operations = {
1256 .d_revalidate = nfs_open_revalidate,
1257 .d_delete = nfs_dentry_delete,
1258 .d_iput = nfs_dentry_iput,
1262 * Use intent information to determine whether we need to substitute
1263 * the NFSv4-style stateful OPEN for the LOOKUP call
1265 static int is_atomic_open(struct nameidata *nd)
1267 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1269 /* NFS does not (yet) have a stateful open for directories */
1270 if (nd->flags & LOOKUP_DIRECTORY)
1272 /* Are we trying to write to a read only partition? */
1273 if (__mnt_is_readonly(nd->path.mnt) &&
1274 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1279 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1281 struct path path = {
1282 .mnt = nd->path.mnt,
1285 struct nfs_open_context *ctx;
1286 struct rpc_cred *cred;
1287 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1289 cred = rpc_lookup_cred();
1291 return ERR_CAST(cred);
1292 ctx = alloc_nfs_open_context(&path, cred, fmode);
1295 return ERR_PTR(-ENOMEM);
1299 static int do_open(struct inode *inode, struct file *filp)
1301 nfs_fscache_set_inode_cookie(inode, filp);
1305 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1310 /* If the open_intent is for execute, we have an extra check to make */
1311 if (ctx->mode & FMODE_EXEC) {
1312 ret = nfs_may_open(ctx->path.dentry->d_inode,
1314 nd->intent.open.flags);
1318 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1320 ret = PTR_ERR(filp);
1322 nfs_file_set_open_context(filp, ctx);
1324 put_nfs_open_context(ctx);
1328 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1330 struct nfs_open_context *ctx;
1332 struct dentry *res = NULL;
1333 struct inode *inode;
1337 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1338 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1340 /* Check that we are indeed trying to open this file */
1341 if (!is_atomic_open(nd))
1344 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1345 res = ERR_PTR(-ENAMETOOLONG);
1348 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1350 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1352 if (nd->flags & LOOKUP_EXCL) {
1353 d_instantiate(dentry, NULL);
1357 ctx = nameidata_to_nfs_open_context(dentry, nd);
1358 res = ERR_CAST(ctx);
1362 open_flags = nd->intent.open.flags;
1363 if (nd->flags & LOOKUP_CREATE) {
1364 attr.ia_mode = nd->intent.open.create_mode;
1365 attr.ia_valid = ATTR_MODE;
1366 if (!IS_POSIXACL(dir))
1367 attr.ia_mode &= ~current_umask();
1369 open_flags &= ~(O_EXCL | O_CREAT);
1373 /* Open the file on the server */
1374 nfs_block_sillyrename(dentry->d_parent);
1375 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1376 if (IS_ERR(inode)) {
1377 nfs_unblock_sillyrename(dentry->d_parent);
1378 put_nfs_open_context(ctx);
1379 switch (PTR_ERR(inode)) {
1380 /* Make a negative dentry */
1382 d_add(dentry, NULL);
1385 /* This turned out not to be a regular file */
1389 if (!(nd->intent.open.flags & O_NOFOLLOW))
1394 res = ERR_CAST(inode);
1398 res = d_add_unique(dentry, inode);
1399 nfs_unblock_sillyrename(dentry->d_parent);
1401 dput(ctx->path.dentry);
1402 ctx->path.dentry = dget(res);
1405 err = nfs_intent_set_file(nd, ctx);
1409 return ERR_PTR(err);
1412 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1415 return nfs_lookup(dir, dentry, nd);
1418 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1420 struct dentry *parent = NULL;
1421 struct inode *inode = dentry->d_inode;
1423 struct nfs_open_context *ctx;
1424 int openflags, ret = 0;
1426 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1429 parent = dget_parent(dentry);
1430 dir = parent->d_inode;
1432 /* We can't create new files in nfs_open_revalidate(), so we
1433 * optimize away revalidation of negative dentries.
1435 if (inode == NULL) {
1436 if (!nfs_neg_need_reval(dir, dentry, nd))
1441 /* NFS only supports OPEN on regular files */
1442 if (!S_ISREG(inode->i_mode))
1444 openflags = nd->intent.open.flags;
1445 /* We cannot do exclusive creation on a positive dentry */
1446 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1448 /* We can't create new files, or truncate existing ones here */
1449 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1451 ctx = nameidata_to_nfs_open_context(dentry, nd);
1456 * Note: we're not holding inode->i_mutex and so may be racing with
1457 * operations that change the directory. We therefore save the
1458 * change attribute *before* we do the RPC call.
1460 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1461 if (IS_ERR(inode)) {
1462 ret = PTR_ERR(inode);
1475 if (inode != dentry->d_inode)
1478 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1479 ret = nfs_intent_set_file(nd, ctx);
1489 put_nfs_open_context(ctx);
1495 return nfs_lookup_revalidate(dentry, nd);
1498 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1499 struct nameidata *nd)
1501 struct nfs_open_context *ctx = NULL;
1506 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1507 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1509 attr.ia_mode = mode;
1510 attr.ia_valid = ATTR_MODE;
1512 if ((nd->flags & LOOKUP_CREATE) != 0) {
1513 open_flags = nd->intent.open.flags;
1515 ctx = nameidata_to_nfs_open_context(dentry, nd);
1516 error = PTR_ERR(ctx);
1521 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1525 error = nfs_intent_set_file(nd, ctx);
1532 put_nfs_open_context(ctx);
1539 #endif /* CONFIG_NFSV4 */
1542 * Code common to create, mkdir, and mknod.
1544 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1545 struct nfs_fattr *fattr)
1547 struct dentry *parent = dget_parent(dentry);
1548 struct inode *dir = parent->d_inode;
1549 struct inode *inode;
1550 int error = -EACCES;
1554 /* We may have been initialized further down */
1555 if (dentry->d_inode)
1557 if (fhandle->size == 0) {
1558 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1562 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1563 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1564 struct nfs_server *server = NFS_SB(dentry->d_sb);
1565 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1569 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1570 error = PTR_ERR(inode);
1573 d_add(dentry, inode);
1578 nfs_mark_for_revalidate(dir);
1584 * Following a failed create operation, we drop the dentry rather
1585 * than retain a negative dentry. This avoids a problem in the event
1586 * that the operation succeeded on the server, but an error in the
1587 * reply path made it appear to have failed.
1589 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1590 struct nameidata *nd)
1595 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1596 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1598 attr.ia_mode = mode;
1599 attr.ia_valid = ATTR_MODE;
1601 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1611 * See comments for nfs_proc_create regarding failed operations.
1614 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1619 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1620 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1622 if (!new_valid_dev(rdev))
1625 attr.ia_mode = mode;
1626 attr.ia_valid = ATTR_MODE;
1628 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1638 * See comments for nfs_proc_create regarding failed operations.
1640 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1645 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1646 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1648 attr.ia_valid = ATTR_MODE;
1649 attr.ia_mode = mode | S_IFDIR;
1651 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1660 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1662 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1666 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1670 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1671 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1673 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1674 /* Ensure the VFS deletes this inode */
1675 if (error == 0 && dentry->d_inode != NULL)
1676 clear_nlink(dentry->d_inode);
1677 else if (error == -ENOENT)
1678 nfs_dentry_handle_enoent(dentry);
1684 * Remove a file after making sure there are no pending writes,
1685 * and after checking that the file has only one user.
1687 * We invalidate the attribute cache and free the inode prior to the operation
1688 * to avoid possible races if the server reuses the inode.
1690 static int nfs_safe_remove(struct dentry *dentry)
1692 struct inode *dir = dentry->d_parent->d_inode;
1693 struct inode *inode = dentry->d_inode;
1696 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1697 dentry->d_parent->d_name.name, dentry->d_name.name);
1699 /* If the dentry was sillyrenamed, we simply call d_delete() */
1700 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1705 if (inode != NULL) {
1706 nfs_inode_return_delegation(inode);
1707 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1708 /* The VFS may want to delete this inode */
1710 nfs_drop_nlink(inode);
1711 nfs_mark_for_revalidate(inode);
1713 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1714 if (error == -ENOENT)
1715 nfs_dentry_handle_enoent(dentry);
1720 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1721 * belongs to an active ".nfs..." file and we return -EBUSY.
1723 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1725 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1728 int need_rehash = 0;
1730 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1731 dir->i_ino, dentry->d_name.name);
1733 spin_lock(&dcache_lock);
1734 spin_lock(&dentry->d_lock);
1735 if (atomic_read(&dentry->d_count) > 1) {
1736 spin_unlock(&dentry->d_lock);
1737 spin_unlock(&dcache_lock);
1738 /* Start asynchronous writeout of the inode */
1739 write_inode_now(dentry->d_inode, 0);
1740 error = nfs_sillyrename(dir, dentry);
1743 if (!d_unhashed(dentry)) {
1747 spin_unlock(&dentry->d_lock);
1748 spin_unlock(&dcache_lock);
1749 error = nfs_safe_remove(dentry);
1750 if (!error || error == -ENOENT) {
1751 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1752 } else if (need_rehash)
1758 * To create a symbolic link, most file systems instantiate a new inode,
1759 * add a page to it containing the path, then write it out to the disk
1760 * using prepare_write/commit_write.
1762 * Unfortunately the NFS client can't create the in-core inode first
1763 * because it needs a file handle to create an in-core inode (see
1764 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1765 * symlink request has completed on the server.
1767 * So instead we allocate a raw page, copy the symname into it, then do
1768 * the SYMLINK request with the page as the buffer. If it succeeds, we
1769 * now have a new file handle and can instantiate an in-core NFS inode
1770 * and move the raw page into its mapping.
1772 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1774 struct pagevec lru_pvec;
1778 unsigned int pathlen = strlen(symname);
1781 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1782 dir->i_ino, dentry->d_name.name, symname);
1784 if (pathlen > PAGE_SIZE)
1785 return -ENAMETOOLONG;
1787 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1788 attr.ia_valid = ATTR_MODE;
1790 page = alloc_page(GFP_HIGHUSER);
1794 kaddr = kmap_atomic(page, KM_USER0);
1795 memcpy(kaddr, symname, pathlen);
1796 if (pathlen < PAGE_SIZE)
1797 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1798 kunmap_atomic(kaddr, KM_USER0);
1800 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1802 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1803 dir->i_sb->s_id, dir->i_ino,
1804 dentry->d_name.name, symname, error);
1811 * No big deal if we can't add this page to the page cache here.
1812 * READLINK will get the missing page from the server if needed.
1814 pagevec_init(&lru_pvec, 0);
1815 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1817 pagevec_add(&lru_pvec, page);
1818 pagevec_lru_add_file(&lru_pvec);
1819 SetPageUptodate(page);
1828 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1830 struct inode *inode = old_dentry->d_inode;
1833 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1834 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1835 dentry->d_parent->d_name.name, dentry->d_name.name);
1837 nfs_inode_return_delegation(inode);
1840 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1843 d_add(dentry, inode);
1850 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1851 * different file handle for the same inode after a rename (e.g. when
1852 * moving to a different directory). A fail-safe method to do so would
1853 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1854 * rename the old file using the sillyrename stuff. This way, the original
1855 * file in old_dir will go away when the last process iput()s the inode.
1859 * It actually works quite well. One needs to have the possibility for
1860 * at least one ".nfs..." file in each directory the file ever gets
1861 * moved or linked to which happens automagically with the new
1862 * implementation that only depends on the dcache stuff instead of
1863 * using the inode layer
1865 * Unfortunately, things are a little more complicated than indicated
1866 * above. For a cross-directory move, we want to make sure we can get
1867 * rid of the old inode after the operation. This means there must be
1868 * no pending writes (if it's a file), and the use count must be 1.
1869 * If these conditions are met, we can drop the dentries before doing
1872 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1873 struct inode *new_dir, struct dentry *new_dentry)
1875 struct inode *old_inode = old_dentry->d_inode;
1876 struct inode *new_inode = new_dentry->d_inode;
1877 struct dentry *dentry = NULL, *rehash = NULL;
1880 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1881 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1882 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1883 atomic_read(&new_dentry->d_count));
1886 * For non-directories, check whether the target is busy and if so,
1887 * make a copy of the dentry and then do a silly-rename. If the
1888 * silly-rename succeeds, the copied dentry is hashed and becomes
1891 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1893 * To prevent any new references to the target during the
1894 * rename, we unhash the dentry in advance.
1896 if (!d_unhashed(new_dentry)) {
1898 rehash = new_dentry;
1901 if (atomic_read(&new_dentry->d_count) > 2) {
1904 /* copy the target dentry's name */
1905 dentry = d_alloc(new_dentry->d_parent,
1906 &new_dentry->d_name);
1910 /* silly-rename the existing target ... */
1911 err = nfs_sillyrename(new_dir, new_dentry);
1915 new_dentry = dentry;
1921 nfs_inode_return_delegation(old_inode);
1922 if (new_inode != NULL)
1923 nfs_inode_return_delegation(new_inode);
1925 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1926 new_dir, &new_dentry->d_name);
1927 nfs_mark_for_revalidate(old_inode);
1932 if (new_inode != NULL)
1933 nfs_drop_nlink(new_inode);
1934 d_move(old_dentry, new_dentry);
1935 nfs_set_verifier(new_dentry,
1936 nfs_save_change_attribute(new_dir));
1937 } else if (error == -ENOENT)
1938 nfs_dentry_handle_enoent(old_dentry);
1940 /* new dentry created? */
1946 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1947 static LIST_HEAD(nfs_access_lru_list);
1948 static atomic_long_t nfs_access_nr_entries;
1950 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1952 put_rpccred(entry->cred);
1954 smp_mb__before_atomic_dec();
1955 atomic_long_dec(&nfs_access_nr_entries);
1956 smp_mb__after_atomic_dec();
1959 static void nfs_access_free_list(struct list_head *head)
1961 struct nfs_access_entry *cache;
1963 while (!list_empty(head)) {
1964 cache = list_entry(head->next, struct nfs_access_entry, lru);
1965 list_del(&cache->lru);
1966 nfs_access_free_entry(cache);
1970 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1973 struct nfs_inode *nfsi, *next;
1974 struct nfs_access_entry *cache;
1976 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1977 return (nr_to_scan == 0) ? 0 : -1;
1979 spin_lock(&nfs_access_lru_lock);
1980 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1981 struct inode *inode;
1983 if (nr_to_scan-- == 0)
1985 inode = &nfsi->vfs_inode;
1986 spin_lock(&inode->i_lock);
1987 if (list_empty(&nfsi->access_cache_entry_lru))
1988 goto remove_lru_entry;
1989 cache = list_entry(nfsi->access_cache_entry_lru.next,
1990 struct nfs_access_entry, lru);
1991 list_move(&cache->lru, &head);
1992 rb_erase(&cache->rb_node, &nfsi->access_cache);
1993 if (!list_empty(&nfsi->access_cache_entry_lru))
1994 list_move_tail(&nfsi->access_cache_inode_lru,
1995 &nfs_access_lru_list);
1998 list_del_init(&nfsi->access_cache_inode_lru);
1999 smp_mb__before_clear_bit();
2000 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2001 smp_mb__after_clear_bit();
2003 spin_unlock(&inode->i_lock);
2005 spin_unlock(&nfs_access_lru_lock);
2006 nfs_access_free_list(&head);
2007 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2010 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2012 struct rb_root *root_node = &nfsi->access_cache;
2014 struct nfs_access_entry *entry;
2016 /* Unhook entries from the cache */
2017 while ((n = rb_first(root_node)) != NULL) {
2018 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2019 rb_erase(n, root_node);
2020 list_move(&entry->lru, head);
2022 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2025 void nfs_access_zap_cache(struct inode *inode)
2029 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2031 /* Remove from global LRU init */
2032 spin_lock(&nfs_access_lru_lock);
2033 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2034 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2036 spin_lock(&inode->i_lock);
2037 __nfs_access_zap_cache(NFS_I(inode), &head);
2038 spin_unlock(&inode->i_lock);
2039 spin_unlock(&nfs_access_lru_lock);
2040 nfs_access_free_list(&head);
2043 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2045 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2046 struct nfs_access_entry *entry;
2049 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2051 if (cred < entry->cred)
2053 else if (cred > entry->cred)
2061 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2063 struct nfs_inode *nfsi = NFS_I(inode);
2064 struct nfs_access_entry *cache;
2067 spin_lock(&inode->i_lock);
2068 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2070 cache = nfs_access_search_rbtree(inode, cred);
2073 if (!nfs_have_delegated_attributes(inode) &&
2074 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2076 res->jiffies = cache->jiffies;
2077 res->cred = cache->cred;
2078 res->mask = cache->mask;
2079 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2082 spin_unlock(&inode->i_lock);
2085 rb_erase(&cache->rb_node, &nfsi->access_cache);
2086 list_del(&cache->lru);
2087 spin_unlock(&inode->i_lock);
2088 nfs_access_free_entry(cache);
2091 spin_unlock(&inode->i_lock);
2092 nfs_access_zap_cache(inode);
2096 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2098 struct nfs_inode *nfsi = NFS_I(inode);
2099 struct rb_root *root_node = &nfsi->access_cache;
2100 struct rb_node **p = &root_node->rb_node;
2101 struct rb_node *parent = NULL;
2102 struct nfs_access_entry *entry;
2104 spin_lock(&inode->i_lock);
2105 while (*p != NULL) {
2107 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2109 if (set->cred < entry->cred)
2110 p = &parent->rb_left;
2111 else if (set->cred > entry->cred)
2112 p = &parent->rb_right;
2116 rb_link_node(&set->rb_node, parent, p);
2117 rb_insert_color(&set->rb_node, root_node);
2118 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2119 spin_unlock(&inode->i_lock);
2122 rb_replace_node(parent, &set->rb_node, root_node);
2123 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2124 list_del(&entry->lru);
2125 spin_unlock(&inode->i_lock);
2126 nfs_access_free_entry(entry);
2129 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2131 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2134 RB_CLEAR_NODE(&cache->rb_node);
2135 cache->jiffies = set->jiffies;
2136 cache->cred = get_rpccred(set->cred);
2137 cache->mask = set->mask;
2139 nfs_access_add_rbtree(inode, cache);
2141 /* Update accounting */
2142 smp_mb__before_atomic_inc();
2143 atomic_long_inc(&nfs_access_nr_entries);
2144 smp_mb__after_atomic_inc();
2146 /* Add inode to global LRU list */
2147 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2148 spin_lock(&nfs_access_lru_lock);
2149 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2150 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2151 &nfs_access_lru_list);
2152 spin_unlock(&nfs_access_lru_lock);
2156 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2158 struct nfs_access_entry cache;
2161 status = nfs_access_get_cached(inode, cred, &cache);
2165 /* Be clever: ask server to check for all possible rights */
2166 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2168 cache.jiffies = jiffies;
2169 status = NFS_PROTO(inode)->access(inode, &cache);
2171 if (status == -ESTALE) {
2172 nfs_zap_caches(inode);
2173 if (!S_ISDIR(inode->i_mode))
2174 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2178 nfs_access_add_cache(inode, &cache);
2180 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2185 static int nfs_open_permission_mask(int openflags)
2189 if (openflags & FMODE_READ)
2191 if (openflags & FMODE_WRITE)
2193 if (openflags & FMODE_EXEC)
2198 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2200 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2203 int nfs_permission(struct inode *inode, int mask)
2205 struct rpc_cred *cred;
2208 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2210 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2212 /* Is this sys_access() ? */
2213 if (mask & (MAY_ACCESS | MAY_CHDIR))
2216 switch (inode->i_mode & S_IFMT) {
2220 /* NFSv4 has atomic_open... */
2221 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2222 && (mask & MAY_OPEN)
2223 && !(mask & MAY_EXEC))
2228 * Optimize away all write operations, since the server
2229 * will check permissions when we perform the op.
2231 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2236 if (!NFS_PROTO(inode)->access)
2239 cred = rpc_lookup_cred();
2240 if (!IS_ERR(cred)) {
2241 res = nfs_do_access(inode, cred, mask);
2244 res = PTR_ERR(cred);
2246 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2249 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2250 inode->i_sb->s_id, inode->i_ino, mask, res);
2253 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2255 res = generic_permission(inode, mask, NULL);
2261 * version-control: t
2262 * kept-new-versions: 5