4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/vmalloc.h>
38 #include "delegation.h"
43 /* #define NFS_DEBUG_VERBOSE 1 */
45 static int nfs_opendir(struct inode *, struct file *);
46 static int nfs_readdir(struct file *, void *, filldir_t);
47 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49 static int nfs_mkdir(struct inode *, struct dentry *, int);
50 static int nfs_rmdir(struct inode *, struct dentry *);
51 static int nfs_unlink(struct inode *, struct dentry *);
52 static int nfs_symlink(struct inode *, struct dentry *, const char *);
53 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55 static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57 static int nfs_fsync_dir(struct file *, int);
58 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 static int nfs_readdir_clear_array(struct page*, gfp_t);
61 const struct file_operations nfs_dir_operations = {
62 .llseek = nfs_llseek_dir,
63 .read = generic_read_dir,
64 .readdir = nfs_readdir,
66 .release = nfs_release,
67 .fsync = nfs_fsync_dir,
70 const struct inode_operations nfs_dir_inode_operations = {
75 .symlink = nfs_symlink,
80 .permission = nfs_permission,
81 .getattr = nfs_getattr,
82 .setattr = nfs_setattr,
85 const struct address_space_operations nfs_dir_addr_space_ops = {
86 .releasepage = nfs_readdir_clear_array,
90 const struct inode_operations nfs3_dir_inode_operations = {
95 .symlink = nfs_symlink,
100 .permission = nfs_permission,
101 .getattr = nfs_getattr,
102 .setattr = nfs_setattr,
103 .listxattr = nfs3_listxattr,
104 .getxattr = nfs3_getxattr,
105 .setxattr = nfs3_setxattr,
106 .removexattr = nfs3_removexattr,
108 #endif /* CONFIG_NFS_V3 */
112 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
113 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
114 const struct inode_operations nfs4_dir_inode_operations = {
115 .create = nfs_open_create,
116 .lookup = nfs_atomic_lookup,
118 .unlink = nfs_unlink,
119 .symlink = nfs_symlink,
123 .rename = nfs_rename,
124 .permission = nfs_permission,
125 .getattr = nfs_getattr,
126 .setattr = nfs_setattr,
127 .getxattr = nfs4_getxattr,
128 .setxattr = nfs4_setxattr,
129 .listxattr = nfs4_listxattr,
132 #endif /* CONFIG_NFS_V4 */
138 nfs_opendir(struct inode *inode, struct file *filp)
142 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
143 filp->f_path.dentry->d_parent->d_name.name,
144 filp->f_path.dentry->d_name.name);
146 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
148 /* Call generic open code in order to cache credentials */
149 res = nfs_open(inode, filp);
150 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
151 /* This is a mountpoint, so d_revalidate will never
152 * have been called, so we need to refresh the
153 * inode (for close-open consistency) ourselves.
155 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
160 struct nfs_cache_array_entry {
166 struct nfs_cache_array {
170 struct nfs_cache_array_entry array[0];
173 #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry))
175 typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
179 unsigned long page_index;
181 loff_t current_index;
182 decode_dirent_t decode;
184 unsigned long timestamp;
185 unsigned long gencount;
186 unsigned int cache_entry_index;
189 } nfs_readdir_descriptor_t;
192 * The caller is responsible for calling nfs_readdir_release_array(page)
195 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
198 return ERR_PTR(-EIO);
199 return (struct nfs_cache_array *)kmap(page);
203 void nfs_readdir_release_array(struct page *page)
209 * we are freeing strings created by nfs_add_to_readdir_array()
212 int nfs_readdir_clear_array(struct page *page, gfp_t mask)
214 struct nfs_cache_array *array = nfs_readdir_get_array(page);
216 for (i = 0; i < array->size; i++)
217 kfree(array->array[i].string.name);
218 nfs_readdir_release_array(page);
223 * the caller is responsible for freeing qstr.name
224 * when called by nfs_readdir_add_to_array, the strings will be freed in
225 * nfs_clear_readdir_array()
228 void nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
231 string->name = kmemdup(name, len, GFP_KERNEL);
232 string->hash = full_name_hash(string->name, string->len);
236 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
238 struct nfs_cache_array *array = nfs_readdir_get_array(page);
240 return PTR_ERR(array);
241 if (array->size >= MAX_READDIR_ARRAY) {
242 nfs_readdir_release_array(page);
246 array->array[array->size].cookie = entry->prev_cookie;
247 array->last_cookie = entry->cookie;
248 array->array[array->size].ino = entry->ino;
249 nfs_readdir_make_qstr(&array->array[array->size].string, entry->name, entry->len);
251 array->eof_index = array->size;
253 nfs_readdir_release_array(page);
258 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
260 loff_t diff = desc->file->f_pos - desc->current_index;
265 if (diff >= array->size) {
266 if (array->eof_index > 0)
268 desc->current_index += array->size;
272 index = (unsigned int)diff;
273 *desc->dir_cookie = array->array[index].cookie;
274 desc->cache_entry_index = index;
275 if (index == array->eof_index)
284 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
287 int status = -EAGAIN;
289 for (i = 0; i < array->size; i++) {
290 if (i == array->eof_index) {
292 status = -EBADCOOKIE;
294 if (array->array[i].cookie == *desc->dir_cookie) {
295 desc->cache_entry_index = i;
305 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
307 struct nfs_cache_array *array;
308 int status = -EBADCOOKIE;
310 if (desc->dir_cookie == NULL)
313 array = nfs_readdir_get_array(desc->page);
315 status = PTR_ERR(array);
319 if (*desc->dir_cookie == 0)
320 status = nfs_readdir_search_for_pos(array, desc);
322 status = nfs_readdir_search_for_cookie(array, desc);
324 nfs_readdir_release_array(desc->page);
329 /* Fill a page with xdr information before transferring to the cache page */
331 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
332 struct nfs_entry *entry, struct file *file, struct inode *inode)
334 struct rpc_cred *cred = nfs_file_cred(file);
335 unsigned long timestamp, gencount;
340 gencount = nfs_inc_attr_generation_counter();
341 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
342 NFS_SERVER(inode)->dtsize, desc->plus);
344 /* We requested READDIRPLUS, but the server doesn't grok it */
345 if (error == -ENOTSUPP && desc->plus) {
346 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
347 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
353 desc->timestamp = timestamp;
354 desc->gencount = gencount;
359 /* Fill in an entry based on the xdr code stored in desc->page */
361 int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream)
363 __be32 *p = desc->decode(stream, entry, desc->plus);
367 entry->fattr->time_start = desc->timestamp;
368 entry->fattr->gencount = desc->gencount;
373 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
375 struct nfs_inode *node;
376 if (dentry->d_inode == NULL)
378 node = NFS_I(dentry->d_inode);
379 if (node->fh.size != entry->fh->size)
381 if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 0)
389 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
391 struct qstr filename;
392 struct dentry *dentry = NULL;
393 struct dentry *alias = NULL;
394 struct inode *dir = parent->d_inode;
397 nfs_readdir_make_qstr(&filename, entry->name, entry->len);
398 if (filename.len == 1 && filename.name[0] == '.')
399 dentry = dget(parent);
400 else if (filename.len == 2 && filename.name[0] == '.'
401 && filename.name[1] == '.')
402 dentry = dget_parent(parent);
404 dentry = d_lookup(parent, &filename);
406 if (dentry != NULL) {
407 if (nfs_same_file(dentry, entry)) {
408 nfs_refresh_inode(dentry->d_inode, entry->fattr);
416 dentry = d_alloc(parent, &filename);
417 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
418 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
422 alias = d_materialise_unique(dentry, inode);
426 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
429 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
433 kfree(filename.name);
437 /* Perform conversion from xdr to cache array */
439 void nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
440 void *xdr_page, struct page *page, unsigned int buflen)
442 struct xdr_stream stream;
444 __be32 *ptr = xdr_page;
446 buf.head->iov_base = xdr_page;
447 buf.head->iov_len = buflen;
448 buf.tail->iov_len = 0;
451 buf.buflen = buf.head->iov_len;
452 buf.len = buf.head->iov_len;
454 xdr_init_decode(&stream, &buf, ptr);
456 while (xdr_decode(desc, entry, &stream) == 0) {
457 if (nfs_readdir_add_to_array(entry, page) == -1)
460 nfs_prime_dcache(desc->file->f_path.dentry, entry);
465 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
468 for (i = 0; i < npages; i++)
473 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
476 vm_unmap_ram(ptr, npages);
477 nfs_readdir_free_pagearray(pages, npages);
481 * nfs_readdir_large_page will allocate pages that must be freed with a call
482 * to nfs_readdir_free_large_page
485 void *nfs_readdir_large_page(struct page **pages, unsigned int npages)
490 for (i = 0; i < npages; i++) {
491 struct page *page = alloc_page(GFP_KERNEL);
497 ptr = vm_map_ram(pages, NFS_MAX_READDIR_PAGES, 0, PAGE_KERNEL);
498 if (!IS_ERR_OR_NULL(ptr))
501 nfs_readdir_free_pagearray(pages, i);
506 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
508 struct page *pages[NFS_MAX_READDIR_PAGES];
509 void *pages_ptr = NULL;
510 struct nfs_entry entry;
511 struct file *file = desc->file;
512 struct nfs_cache_array *array;
514 unsigned int array_size = ARRAY_SIZE(pages);
516 entry.prev_cookie = 0;
517 entry.cookie = *desc->dir_cookie;
519 entry.fh = nfs_alloc_fhandle();
520 entry.fattr = nfs_alloc_fattr();
521 if (entry.fh == NULL || entry.fattr == NULL)
524 array = nfs_readdir_get_array(page);
525 memset(array, 0, sizeof(struct nfs_cache_array));
526 array->eof_index = -1;
528 pages_ptr = nfs_readdir_large_page(pages, array_size);
530 goto out_release_array;
532 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
536 nfs_readdir_page_filler(desc, &entry, pages_ptr, page, array_size * PAGE_SIZE);
537 } while (array->eof_index < 0 && array->size < MAX_READDIR_ARRAY);
539 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
541 nfs_readdir_release_array(page);
543 nfs_free_fattr(entry.fattr);
544 nfs_free_fhandle(entry.fh);
549 * Now we cache directories properly, by converting xdr information
550 * to an array that can be used for lookups later. This results in
551 * fewer cache pages, since we can store more information on each page.
552 * We only need to convert from xdr once so future lookups are much simpler
555 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
557 struct inode *inode = desc->file->f_path.dentry->d_inode;
559 if (nfs_readdir_xdr_to_array(desc, page, inode) == -1)
561 SetPageUptodate(page);
563 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
564 /* Should never happen */
565 nfs_zap_mapping(inode, inode->i_mapping);
575 void cache_page_release(nfs_readdir_descriptor_t *desc)
577 page_cache_release(desc->page);
582 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
585 page = read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
586 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
593 * Returns 0 if desc->dir_cookie was found on page desc->page_index
596 int find_cache_page(nfs_readdir_descriptor_t *desc)
600 desc->page = get_cache_page(desc);
601 if (IS_ERR(desc->page))
602 return PTR_ERR(desc->page);
604 res = nfs_readdir_search_array(desc);
607 cache_page_release(desc);
611 /* Search for desc->dir_cookie from the beginning of the page cache */
613 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
618 res = find_cache_page(desc);
626 static inline unsigned int dt_type(struct inode *inode)
628 return (inode->i_mode >> 12) & 15;
632 * Once we've found the start of the dirent within a page: fill 'er up...
635 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
638 struct file *file = desc->file;
641 struct nfs_cache_array *array = NULL;
642 unsigned int d_type = DT_UNKNOWN;
643 struct dentry *dentry = NULL;
645 array = nfs_readdir_get_array(desc->page);
647 for (i = desc->cache_entry_index; i < array->size; i++) {
650 res = filldir(dirent, array->array[i].string.name,
651 array->array[i].string.len, file->f_pos,
652 nfs_compat_user_ino64(array->array[i].ino), d_type);
656 desc->cache_entry_index = i;
657 if (i < (array->size-1))
658 *desc->dir_cookie = array->array[i+1].cookie;
660 *desc->dir_cookie = array->last_cookie;
661 if (i == array->eof_index) {
667 nfs_readdir_release_array(desc->page);
668 cache_page_release(desc);
671 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
672 (unsigned long long)*desc->dir_cookie, res);
677 * If we cannot find a cookie in our cache, we suspect that this is
678 * because it points to a deleted file, so we ask the server to return
679 * whatever it thinks is the next entry. We then feed this to filldir.
680 * If all goes well, we should then be able to find our way round the
681 * cache on the next call to readdir_search_pagecache();
683 * NOTE: we cannot add the anonymous page to the pagecache because
684 * the data it contains might not be page aligned. Besides,
685 * we should already have a complete representation of the
686 * directory in the page cache by the time we get here.
689 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
692 struct page *page = NULL;
694 struct inode *inode = desc->file->f_path.dentry->d_inode;
696 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
697 (unsigned long long)*desc->dir_cookie);
699 page = alloc_page(GFP_HIGHUSER);
705 if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) {
710 desc->page_index = 0;
712 status = nfs_do_filldir(desc, dirent, filldir);
715 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
719 cache_page_release(desc);
723 /* The file offset position represents the dirent entry number. A
724 last cookie cache takes care of the common case of reading the
727 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
729 struct dentry *dentry = filp->f_path.dentry;
730 struct inode *inode = dentry->d_inode;
731 nfs_readdir_descriptor_t my_desc,
735 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
736 dentry->d_parent->d_name.name, dentry->d_name.name,
737 (long long)filp->f_pos);
738 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
741 * filp->f_pos points to the dirent entry number.
742 * *desc->dir_cookie has the cookie for the next entry. We have
743 * to either find the entry with the appropriate number or
744 * revalidate the cookie.
746 memset(desc, 0, sizeof(*desc));
749 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
750 desc->decode = NFS_PROTO(inode)->decode_dirent;
751 desc->plus = NFS_USE_READDIRPLUS(inode);
753 nfs_block_sillyrename(dentry);
754 res = nfs_revalidate_mapping(inode, filp->f_mapping);
758 while (desc->eof != 1) {
759 res = readdir_search_pagecache(desc);
761 if (res == -EBADCOOKIE) {
762 /* This means either end of directory */
763 if (*desc->dir_cookie && desc->eof == 0) {
764 /* Or that the server has 'lost' a cookie */
765 res = uncached_readdir(desc, dirent, filldir);
772 if (res == -ETOOSMALL && desc->plus) {
773 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
774 nfs_zap_caches(inode);
775 desc->page_index = 0;
783 res = nfs_do_filldir(desc, dirent, filldir);
790 nfs_unblock_sillyrename(dentry);
793 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
794 dentry->d_parent->d_name.name, dentry->d_name.name,
799 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
801 struct dentry *dentry = filp->f_path.dentry;
802 struct inode *inode = dentry->d_inode;
804 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
805 dentry->d_parent->d_name.name,
809 mutex_lock(&inode->i_mutex);
812 offset += filp->f_pos;
820 if (offset != filp->f_pos) {
821 filp->f_pos = offset;
822 nfs_file_open_context(filp)->dir_cookie = 0;
825 mutex_unlock(&inode->i_mutex);
830 * All directory operations under NFS are synchronous, so fsync()
831 * is a dummy operation.
833 static int nfs_fsync_dir(struct file *filp, int datasync)
835 struct dentry *dentry = filp->f_path.dentry;
837 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
838 dentry->d_parent->d_name.name, dentry->d_name.name,
841 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
846 * nfs_force_lookup_revalidate - Mark the directory as having changed
847 * @dir - pointer to directory inode
849 * This forces the revalidation code in nfs_lookup_revalidate() to do a
850 * full lookup on all child dentries of 'dir' whenever a change occurs
851 * on the server that might have invalidated our dcache.
853 * The caller should be holding dir->i_lock
855 void nfs_force_lookup_revalidate(struct inode *dir)
857 NFS_I(dir)->cache_change_attribute++;
861 * A check for whether or not the parent directory has changed.
862 * In the case it has, we assume that the dentries are untrustworthy
863 * and may need to be looked up again.
865 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
869 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
871 if (!nfs_verify_change_attribute(dir, dentry->d_time))
873 /* Revalidate nfsi->cache_change_attribute before we declare a match */
874 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
876 if (!nfs_verify_change_attribute(dir, dentry->d_time))
882 * Return the intent data that applies to this particular path component
884 * Note that the current set of intents only apply to the very last
885 * component of the path.
886 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
888 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
890 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
892 return nd->flags & mask;
896 * Use intent information to check whether or not we're going to do
897 * an O_EXCL create using this path component.
899 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
901 if (NFS_PROTO(dir)->version == 2)
903 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
907 * Inode and filehandle revalidation for lookups.
909 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
910 * or if the intent information indicates that we're about to open this
911 * particular file and the "nocto" mount flag is not set.
915 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
917 struct nfs_server *server = NFS_SERVER(inode);
919 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
922 /* VFS wants an on-the-wire revalidation */
923 if (nd->flags & LOOKUP_REVAL)
925 /* This is an open(2) */
926 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
927 !(server->flags & NFS_MOUNT_NOCTO) &&
928 (S_ISREG(inode->i_mode) ||
929 S_ISDIR(inode->i_mode)))
933 return nfs_revalidate_inode(server, inode);
935 return __nfs_revalidate_inode(server, inode);
939 * We judge how long we want to trust negative
940 * dentries by looking at the parent inode mtime.
942 * If parent mtime has changed, we revalidate, else we wait for a
943 * period corresponding to the parent's attribute cache timeout value.
946 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
947 struct nameidata *nd)
949 /* Don't revalidate a negative dentry if we're creating a new file */
950 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
952 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
954 return !nfs_check_verifier(dir, dentry);
958 * This is called every time the dcache has a lookup hit,
959 * and we should check whether we can really trust that
962 * NOTE! The hit can be a negative hit too, don't assume
965 * If the parent directory is seen to have changed, we throw out the
966 * cached dentry and do a new lookup.
968 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
972 struct dentry *parent;
973 struct nfs_fh *fhandle = NULL;
974 struct nfs_fattr *fattr = NULL;
977 parent = dget_parent(dentry);
978 dir = parent->d_inode;
979 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
980 inode = dentry->d_inode;
983 if (nfs_neg_need_reval(dir, dentry, nd))
988 if (is_bad_inode(inode)) {
989 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
990 __func__, dentry->d_parent->d_name.name,
991 dentry->d_name.name);
995 if (nfs_have_delegation(inode, FMODE_READ))
996 goto out_set_verifier;
998 /* Force a full look up iff the parent directory has changed */
999 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1000 if (nfs_lookup_verify_inode(inode, nd))
1001 goto out_zap_parent;
1005 if (NFS_STALE(inode))
1009 fhandle = nfs_alloc_fhandle();
1010 fattr = nfs_alloc_fattr();
1011 if (fhandle == NULL || fattr == NULL)
1014 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1017 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1019 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1022 nfs_free_fattr(fattr);
1023 nfs_free_fhandle(fhandle);
1025 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1028 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1029 __func__, dentry->d_parent->d_name.name,
1030 dentry->d_name.name);
1033 nfs_zap_caches(dir);
1035 nfs_mark_for_revalidate(dir);
1036 if (inode && S_ISDIR(inode->i_mode)) {
1037 /* Purge readdir caches. */
1038 nfs_zap_caches(inode);
1039 /* If we have submounts, don't unhash ! */
1040 if (have_submounts(dentry))
1042 if (dentry->d_flags & DCACHE_DISCONNECTED)
1044 shrink_dcache_parent(dentry);
1047 nfs_free_fattr(fattr);
1048 nfs_free_fhandle(fhandle);
1050 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1051 __func__, dentry->d_parent->d_name.name,
1052 dentry->d_name.name);
1055 nfs_free_fattr(fattr);
1056 nfs_free_fhandle(fhandle);
1058 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1059 __func__, dentry->d_parent->d_name.name,
1060 dentry->d_name.name, error);
1065 * This is called from dput() when d_count is going to 0.
1067 static int nfs_dentry_delete(struct dentry *dentry)
1069 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1070 dentry->d_parent->d_name.name, dentry->d_name.name,
1073 /* Unhash any dentry with a stale inode */
1074 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1077 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1078 /* Unhash it, so that ->d_iput() would be called */
1081 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1082 /* Unhash it, so that ancestors of killed async unlink
1083 * files will be cleaned up during umount */
1090 static void nfs_drop_nlink(struct inode *inode)
1092 spin_lock(&inode->i_lock);
1093 if (inode->i_nlink > 0)
1095 spin_unlock(&inode->i_lock);
1099 * Called when the dentry loses inode.
1100 * We use it to clean up silly-renamed files.
1102 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1104 if (S_ISDIR(inode->i_mode))
1105 /* drop any readdir cache as it could easily be old */
1106 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1108 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1110 nfs_complete_unlink(dentry, inode);
1115 const struct dentry_operations nfs_dentry_operations = {
1116 .d_revalidate = nfs_lookup_revalidate,
1117 .d_delete = nfs_dentry_delete,
1118 .d_iput = nfs_dentry_iput,
1121 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1124 struct dentry *parent;
1125 struct inode *inode = NULL;
1126 struct nfs_fh *fhandle = NULL;
1127 struct nfs_fattr *fattr = NULL;
1130 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1131 dentry->d_parent->d_name.name, dentry->d_name.name);
1132 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1134 res = ERR_PTR(-ENAMETOOLONG);
1135 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1138 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1141 * If we're doing an exclusive create, optimize away the lookup
1142 * but don't hash the dentry.
1144 if (nfs_is_exclusive_create(dir, nd)) {
1145 d_instantiate(dentry, NULL);
1150 res = ERR_PTR(-ENOMEM);
1151 fhandle = nfs_alloc_fhandle();
1152 fattr = nfs_alloc_fattr();
1153 if (fhandle == NULL || fattr == NULL)
1156 parent = dentry->d_parent;
1157 /* Protect against concurrent sillydeletes */
1158 nfs_block_sillyrename(parent);
1159 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1160 if (error == -ENOENT)
1163 res = ERR_PTR(error);
1164 goto out_unblock_sillyrename;
1166 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1167 res = (struct dentry *)inode;
1169 goto out_unblock_sillyrename;
1172 res = d_materialise_unique(dentry, inode);
1175 goto out_unblock_sillyrename;
1178 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1179 out_unblock_sillyrename:
1180 nfs_unblock_sillyrename(parent);
1182 nfs_free_fattr(fattr);
1183 nfs_free_fhandle(fhandle);
1187 #ifdef CONFIG_NFS_V4
1188 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1190 const struct dentry_operations nfs4_dentry_operations = {
1191 .d_revalidate = nfs_open_revalidate,
1192 .d_delete = nfs_dentry_delete,
1193 .d_iput = nfs_dentry_iput,
1197 * Use intent information to determine whether we need to substitute
1198 * the NFSv4-style stateful OPEN for the LOOKUP call
1200 static int is_atomic_open(struct nameidata *nd)
1202 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1204 /* NFS does not (yet) have a stateful open for directories */
1205 if (nd->flags & LOOKUP_DIRECTORY)
1207 /* Are we trying to write to a read only partition? */
1208 if (__mnt_is_readonly(nd->path.mnt) &&
1209 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1214 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1216 struct path path = {
1217 .mnt = nd->path.mnt,
1220 struct nfs_open_context *ctx;
1221 struct rpc_cred *cred;
1222 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1224 cred = rpc_lookup_cred();
1226 return ERR_CAST(cred);
1227 ctx = alloc_nfs_open_context(&path, cred, fmode);
1230 return ERR_PTR(-ENOMEM);
1234 static int do_open(struct inode *inode, struct file *filp)
1236 nfs_fscache_set_inode_cookie(inode, filp);
1240 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1245 /* If the open_intent is for execute, we have an extra check to make */
1246 if (ctx->mode & FMODE_EXEC) {
1247 ret = nfs_may_open(ctx->path.dentry->d_inode,
1249 nd->intent.open.flags);
1253 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1255 ret = PTR_ERR(filp);
1257 nfs_file_set_open_context(filp, ctx);
1259 put_nfs_open_context(ctx);
1263 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1265 struct nfs_open_context *ctx;
1267 struct dentry *res = NULL;
1268 struct inode *inode;
1272 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1273 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1275 /* Check that we are indeed trying to open this file */
1276 if (!is_atomic_open(nd))
1279 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1280 res = ERR_PTR(-ENAMETOOLONG);
1283 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1285 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1287 if (nd->flags & LOOKUP_EXCL) {
1288 d_instantiate(dentry, NULL);
1292 ctx = nameidata_to_nfs_open_context(dentry, nd);
1293 res = ERR_CAST(ctx);
1297 open_flags = nd->intent.open.flags;
1298 if (nd->flags & LOOKUP_CREATE) {
1299 attr.ia_mode = nd->intent.open.create_mode;
1300 attr.ia_valid = ATTR_MODE;
1301 if (!IS_POSIXACL(dir))
1302 attr.ia_mode &= ~current_umask();
1304 open_flags &= ~(O_EXCL | O_CREAT);
1308 /* Open the file on the server */
1309 nfs_block_sillyrename(dentry->d_parent);
1310 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1311 if (IS_ERR(inode)) {
1312 nfs_unblock_sillyrename(dentry->d_parent);
1313 put_nfs_open_context(ctx);
1314 switch (PTR_ERR(inode)) {
1315 /* Make a negative dentry */
1317 d_add(dentry, NULL);
1320 /* This turned out not to be a regular file */
1325 if (!(nd->intent.open.flags & O_NOFOLLOW))
1329 res = ERR_CAST(inode);
1333 res = d_add_unique(dentry, inode);
1334 nfs_unblock_sillyrename(dentry->d_parent);
1336 dput(ctx->path.dentry);
1337 ctx->path.dentry = dget(res);
1340 err = nfs_intent_set_file(nd, ctx);
1344 return ERR_PTR(err);
1347 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1350 return nfs_lookup(dir, dentry, nd);
1353 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1355 struct dentry *parent = NULL;
1356 struct inode *inode = dentry->d_inode;
1358 struct nfs_open_context *ctx;
1359 int openflags, ret = 0;
1361 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1364 parent = dget_parent(dentry);
1365 dir = parent->d_inode;
1367 /* We can't create new files in nfs_open_revalidate(), so we
1368 * optimize away revalidation of negative dentries.
1370 if (inode == NULL) {
1371 if (!nfs_neg_need_reval(dir, dentry, nd))
1376 /* NFS only supports OPEN on regular files */
1377 if (!S_ISREG(inode->i_mode))
1379 openflags = nd->intent.open.flags;
1380 /* We cannot do exclusive creation on a positive dentry */
1381 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1383 /* We can't create new files, or truncate existing ones here */
1384 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1386 ctx = nameidata_to_nfs_open_context(dentry, nd);
1391 * Note: we're not holding inode->i_mutex and so may be racing with
1392 * operations that change the directory. We therefore save the
1393 * change attribute *before* we do the RPC call.
1395 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1396 if (IS_ERR(inode)) {
1397 ret = PTR_ERR(inode);
1410 if (inode != dentry->d_inode)
1413 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1414 ret = nfs_intent_set_file(nd, ctx);
1424 put_nfs_open_context(ctx);
1430 return nfs_lookup_revalidate(dentry, nd);
1433 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1434 struct nameidata *nd)
1436 struct nfs_open_context *ctx = NULL;
1441 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1442 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1444 attr.ia_mode = mode;
1445 attr.ia_valid = ATTR_MODE;
1447 if ((nd->flags & LOOKUP_CREATE) != 0) {
1448 open_flags = nd->intent.open.flags;
1450 ctx = nameidata_to_nfs_open_context(dentry, nd);
1451 error = PTR_ERR(ctx);
1456 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1460 error = nfs_intent_set_file(nd, ctx);
1467 put_nfs_open_context(ctx);
1474 #endif /* CONFIG_NFSV4 */
1477 * Code common to create, mkdir, and mknod.
1479 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1480 struct nfs_fattr *fattr)
1482 struct dentry *parent = dget_parent(dentry);
1483 struct inode *dir = parent->d_inode;
1484 struct inode *inode;
1485 int error = -EACCES;
1489 /* We may have been initialized further down */
1490 if (dentry->d_inode)
1492 if (fhandle->size == 0) {
1493 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1497 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1498 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1499 struct nfs_server *server = NFS_SB(dentry->d_sb);
1500 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1504 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1505 error = PTR_ERR(inode);
1508 d_add(dentry, inode);
1513 nfs_mark_for_revalidate(dir);
1519 * Following a failed create operation, we drop the dentry rather
1520 * than retain a negative dentry. This avoids a problem in the event
1521 * that the operation succeeded on the server, but an error in the
1522 * reply path made it appear to have failed.
1524 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1525 struct nameidata *nd)
1530 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1531 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1533 attr.ia_mode = mode;
1534 attr.ia_valid = ATTR_MODE;
1536 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1546 * See comments for nfs_proc_create regarding failed operations.
1549 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1554 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1555 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1557 if (!new_valid_dev(rdev))
1560 attr.ia_mode = mode;
1561 attr.ia_valid = ATTR_MODE;
1563 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1573 * See comments for nfs_proc_create regarding failed operations.
1575 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1580 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1581 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1583 attr.ia_valid = ATTR_MODE;
1584 attr.ia_mode = mode | S_IFDIR;
1586 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1595 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1597 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1601 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1605 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1606 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1608 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1609 /* Ensure the VFS deletes this inode */
1610 if (error == 0 && dentry->d_inode != NULL)
1611 clear_nlink(dentry->d_inode);
1612 else if (error == -ENOENT)
1613 nfs_dentry_handle_enoent(dentry);
1619 * Remove a file after making sure there are no pending writes,
1620 * and after checking that the file has only one user.
1622 * We invalidate the attribute cache and free the inode prior to the operation
1623 * to avoid possible races if the server reuses the inode.
1625 static int nfs_safe_remove(struct dentry *dentry)
1627 struct inode *dir = dentry->d_parent->d_inode;
1628 struct inode *inode = dentry->d_inode;
1631 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1632 dentry->d_parent->d_name.name, dentry->d_name.name);
1634 /* If the dentry was sillyrenamed, we simply call d_delete() */
1635 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1640 if (inode != NULL) {
1641 nfs_inode_return_delegation(inode);
1642 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1643 /* The VFS may want to delete this inode */
1645 nfs_drop_nlink(inode);
1646 nfs_mark_for_revalidate(inode);
1648 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1649 if (error == -ENOENT)
1650 nfs_dentry_handle_enoent(dentry);
1655 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1656 * belongs to an active ".nfs..." file and we return -EBUSY.
1658 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1660 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1663 int need_rehash = 0;
1665 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1666 dir->i_ino, dentry->d_name.name);
1668 spin_lock(&dcache_lock);
1669 spin_lock(&dentry->d_lock);
1670 if (atomic_read(&dentry->d_count) > 1) {
1671 spin_unlock(&dentry->d_lock);
1672 spin_unlock(&dcache_lock);
1673 /* Start asynchronous writeout of the inode */
1674 write_inode_now(dentry->d_inode, 0);
1675 error = nfs_sillyrename(dir, dentry);
1678 if (!d_unhashed(dentry)) {
1682 spin_unlock(&dentry->d_lock);
1683 spin_unlock(&dcache_lock);
1684 error = nfs_safe_remove(dentry);
1685 if (!error || error == -ENOENT) {
1686 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1687 } else if (need_rehash)
1693 * To create a symbolic link, most file systems instantiate a new inode,
1694 * add a page to it containing the path, then write it out to the disk
1695 * using prepare_write/commit_write.
1697 * Unfortunately the NFS client can't create the in-core inode first
1698 * because it needs a file handle to create an in-core inode (see
1699 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1700 * symlink request has completed on the server.
1702 * So instead we allocate a raw page, copy the symname into it, then do
1703 * the SYMLINK request with the page as the buffer. If it succeeds, we
1704 * now have a new file handle and can instantiate an in-core NFS inode
1705 * and move the raw page into its mapping.
1707 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1709 struct pagevec lru_pvec;
1713 unsigned int pathlen = strlen(symname);
1716 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1717 dir->i_ino, dentry->d_name.name, symname);
1719 if (pathlen > PAGE_SIZE)
1720 return -ENAMETOOLONG;
1722 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1723 attr.ia_valid = ATTR_MODE;
1725 page = alloc_page(GFP_HIGHUSER);
1729 kaddr = kmap_atomic(page, KM_USER0);
1730 memcpy(kaddr, symname, pathlen);
1731 if (pathlen < PAGE_SIZE)
1732 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1733 kunmap_atomic(kaddr, KM_USER0);
1735 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1737 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1738 dir->i_sb->s_id, dir->i_ino,
1739 dentry->d_name.name, symname, error);
1746 * No big deal if we can't add this page to the page cache here.
1747 * READLINK will get the missing page from the server if needed.
1749 pagevec_init(&lru_pvec, 0);
1750 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1752 pagevec_add(&lru_pvec, page);
1753 pagevec_lru_add_file(&lru_pvec);
1754 SetPageUptodate(page);
1763 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1765 struct inode *inode = old_dentry->d_inode;
1768 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1769 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1770 dentry->d_parent->d_name.name, dentry->d_name.name);
1772 nfs_inode_return_delegation(inode);
1775 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1777 atomic_inc(&inode->i_count);
1778 d_add(dentry, inode);
1785 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1786 * different file handle for the same inode after a rename (e.g. when
1787 * moving to a different directory). A fail-safe method to do so would
1788 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1789 * rename the old file using the sillyrename stuff. This way, the original
1790 * file in old_dir will go away when the last process iput()s the inode.
1794 * It actually works quite well. One needs to have the possibility for
1795 * at least one ".nfs..." file in each directory the file ever gets
1796 * moved or linked to which happens automagically with the new
1797 * implementation that only depends on the dcache stuff instead of
1798 * using the inode layer
1800 * Unfortunately, things are a little more complicated than indicated
1801 * above. For a cross-directory move, we want to make sure we can get
1802 * rid of the old inode after the operation. This means there must be
1803 * no pending writes (if it's a file), and the use count must be 1.
1804 * If these conditions are met, we can drop the dentries before doing
1807 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1808 struct inode *new_dir, struct dentry *new_dentry)
1810 struct inode *old_inode = old_dentry->d_inode;
1811 struct inode *new_inode = new_dentry->d_inode;
1812 struct dentry *dentry = NULL, *rehash = NULL;
1815 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1816 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1817 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1818 atomic_read(&new_dentry->d_count));
1821 * For non-directories, check whether the target is busy and if so,
1822 * make a copy of the dentry and then do a silly-rename. If the
1823 * silly-rename succeeds, the copied dentry is hashed and becomes
1826 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1828 * To prevent any new references to the target during the
1829 * rename, we unhash the dentry in advance.
1831 if (!d_unhashed(new_dentry)) {
1833 rehash = new_dentry;
1836 if (atomic_read(&new_dentry->d_count) > 2) {
1839 /* copy the target dentry's name */
1840 dentry = d_alloc(new_dentry->d_parent,
1841 &new_dentry->d_name);
1845 /* silly-rename the existing target ... */
1846 err = nfs_sillyrename(new_dir, new_dentry);
1850 new_dentry = dentry;
1856 nfs_inode_return_delegation(old_inode);
1857 if (new_inode != NULL)
1858 nfs_inode_return_delegation(new_inode);
1860 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1861 new_dir, &new_dentry->d_name);
1862 nfs_mark_for_revalidate(old_inode);
1867 if (new_inode != NULL)
1868 nfs_drop_nlink(new_inode);
1869 d_move(old_dentry, new_dentry);
1870 nfs_set_verifier(new_dentry,
1871 nfs_save_change_attribute(new_dir));
1872 } else if (error == -ENOENT)
1873 nfs_dentry_handle_enoent(old_dentry);
1875 /* new dentry created? */
1881 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1882 static LIST_HEAD(nfs_access_lru_list);
1883 static atomic_long_t nfs_access_nr_entries;
1885 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1887 put_rpccred(entry->cred);
1889 smp_mb__before_atomic_dec();
1890 atomic_long_dec(&nfs_access_nr_entries);
1891 smp_mb__after_atomic_dec();
1894 static void nfs_access_free_list(struct list_head *head)
1896 struct nfs_access_entry *cache;
1898 while (!list_empty(head)) {
1899 cache = list_entry(head->next, struct nfs_access_entry, lru);
1900 list_del(&cache->lru);
1901 nfs_access_free_entry(cache);
1905 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1908 struct nfs_inode *nfsi, *next;
1909 struct nfs_access_entry *cache;
1911 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1912 return (nr_to_scan == 0) ? 0 : -1;
1914 spin_lock(&nfs_access_lru_lock);
1915 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1916 struct inode *inode;
1918 if (nr_to_scan-- == 0)
1920 inode = &nfsi->vfs_inode;
1921 spin_lock(&inode->i_lock);
1922 if (list_empty(&nfsi->access_cache_entry_lru))
1923 goto remove_lru_entry;
1924 cache = list_entry(nfsi->access_cache_entry_lru.next,
1925 struct nfs_access_entry, lru);
1926 list_move(&cache->lru, &head);
1927 rb_erase(&cache->rb_node, &nfsi->access_cache);
1928 if (!list_empty(&nfsi->access_cache_entry_lru))
1929 list_move_tail(&nfsi->access_cache_inode_lru,
1930 &nfs_access_lru_list);
1933 list_del_init(&nfsi->access_cache_inode_lru);
1934 smp_mb__before_clear_bit();
1935 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1936 smp_mb__after_clear_bit();
1938 spin_unlock(&inode->i_lock);
1940 spin_unlock(&nfs_access_lru_lock);
1941 nfs_access_free_list(&head);
1942 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1945 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1947 struct rb_root *root_node = &nfsi->access_cache;
1949 struct nfs_access_entry *entry;
1951 /* Unhook entries from the cache */
1952 while ((n = rb_first(root_node)) != NULL) {
1953 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1954 rb_erase(n, root_node);
1955 list_move(&entry->lru, head);
1957 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1960 void nfs_access_zap_cache(struct inode *inode)
1964 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
1966 /* Remove from global LRU init */
1967 spin_lock(&nfs_access_lru_lock);
1968 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
1969 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1971 spin_lock(&inode->i_lock);
1972 __nfs_access_zap_cache(NFS_I(inode), &head);
1973 spin_unlock(&inode->i_lock);
1974 spin_unlock(&nfs_access_lru_lock);
1975 nfs_access_free_list(&head);
1978 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1980 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1981 struct nfs_access_entry *entry;
1984 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1986 if (cred < entry->cred)
1988 else if (cred > entry->cred)
1996 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1998 struct nfs_inode *nfsi = NFS_I(inode);
1999 struct nfs_access_entry *cache;
2002 spin_lock(&inode->i_lock);
2003 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2005 cache = nfs_access_search_rbtree(inode, cred);
2008 if (!nfs_have_delegated_attributes(inode) &&
2009 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2011 res->jiffies = cache->jiffies;
2012 res->cred = cache->cred;
2013 res->mask = cache->mask;
2014 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2017 spin_unlock(&inode->i_lock);
2020 rb_erase(&cache->rb_node, &nfsi->access_cache);
2021 list_del(&cache->lru);
2022 spin_unlock(&inode->i_lock);
2023 nfs_access_free_entry(cache);
2026 spin_unlock(&inode->i_lock);
2027 nfs_access_zap_cache(inode);
2031 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2033 struct nfs_inode *nfsi = NFS_I(inode);
2034 struct rb_root *root_node = &nfsi->access_cache;
2035 struct rb_node **p = &root_node->rb_node;
2036 struct rb_node *parent = NULL;
2037 struct nfs_access_entry *entry;
2039 spin_lock(&inode->i_lock);
2040 while (*p != NULL) {
2042 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2044 if (set->cred < entry->cred)
2045 p = &parent->rb_left;
2046 else if (set->cred > entry->cred)
2047 p = &parent->rb_right;
2051 rb_link_node(&set->rb_node, parent, p);
2052 rb_insert_color(&set->rb_node, root_node);
2053 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2054 spin_unlock(&inode->i_lock);
2057 rb_replace_node(parent, &set->rb_node, root_node);
2058 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2059 list_del(&entry->lru);
2060 spin_unlock(&inode->i_lock);
2061 nfs_access_free_entry(entry);
2064 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2066 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2069 RB_CLEAR_NODE(&cache->rb_node);
2070 cache->jiffies = set->jiffies;
2071 cache->cred = get_rpccred(set->cred);
2072 cache->mask = set->mask;
2074 nfs_access_add_rbtree(inode, cache);
2076 /* Update accounting */
2077 smp_mb__before_atomic_inc();
2078 atomic_long_inc(&nfs_access_nr_entries);
2079 smp_mb__after_atomic_inc();
2081 /* Add inode to global LRU list */
2082 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2083 spin_lock(&nfs_access_lru_lock);
2084 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2085 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2086 &nfs_access_lru_list);
2087 spin_unlock(&nfs_access_lru_lock);
2091 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2093 struct nfs_access_entry cache;
2096 status = nfs_access_get_cached(inode, cred, &cache);
2100 /* Be clever: ask server to check for all possible rights */
2101 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2103 cache.jiffies = jiffies;
2104 status = NFS_PROTO(inode)->access(inode, &cache);
2106 if (status == -ESTALE) {
2107 nfs_zap_caches(inode);
2108 if (!S_ISDIR(inode->i_mode))
2109 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2113 nfs_access_add_cache(inode, &cache);
2115 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2120 static int nfs_open_permission_mask(int openflags)
2124 if (openflags & FMODE_READ)
2126 if (openflags & FMODE_WRITE)
2128 if (openflags & FMODE_EXEC)
2133 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2135 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2138 int nfs_permission(struct inode *inode, int mask)
2140 struct rpc_cred *cred;
2143 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2145 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2147 /* Is this sys_access() ? */
2148 if (mask & (MAY_ACCESS | MAY_CHDIR))
2151 switch (inode->i_mode & S_IFMT) {
2155 /* NFSv4 has atomic_open... */
2156 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2157 && (mask & MAY_OPEN)
2158 && !(mask & MAY_EXEC))
2163 * Optimize away all write operations, since the server
2164 * will check permissions when we perform the op.
2166 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2171 if (!NFS_PROTO(inode)->access)
2174 cred = rpc_lookup_cred();
2175 if (!IS_ERR(cred)) {
2176 res = nfs_do_access(inode, cred, mask);
2179 res = PTR_ERR(cred);
2181 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2184 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2185 inode->i_sb->s_id, inode->i_ino, mask, res);
2188 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2190 res = generic_permission(inode, mask, NULL);
2196 * version-control: t
2197 * kept-new-versions: 5