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/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
50 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
51 static void nfs_readdir_clear_array(struct page*);
53 const struct file_operations nfs_dir_operations = {
54 .llseek = nfs_llseek_dir,
55 .read = generic_read_dir,
56 .readdir = nfs_readdir,
58 .release = nfs_closedir,
59 .fsync = nfs_fsync_dir,
62 const struct address_space_operations nfs_dir_aops = {
63 .freepage = nfs_readdir_clear_array,
66 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
68 struct nfs_open_dir_context *ctx;
69 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
72 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
75 ctx->cred = get_rpccred(cred);
78 return ERR_PTR(-ENOMEM);
81 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
83 put_rpccred(ctx->cred);
91 nfs_opendir(struct inode *inode, struct file *filp)
94 struct nfs_open_dir_context *ctx;
95 struct rpc_cred *cred;
97 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
98 filp->f_path.dentry->d_parent->d_name.name,
99 filp->f_path.dentry->d_name.name);
101 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
103 cred = rpc_lookup_cred();
105 return PTR_ERR(cred);
106 ctx = alloc_nfs_open_dir_context(inode, cred);
111 filp->private_data = ctx;
112 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
113 /* This is a mountpoint, so d_revalidate will never
114 * have been called, so we need to refresh the
115 * inode (for close-open consistency) ourselves.
117 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
125 nfs_closedir(struct inode *inode, struct file *filp)
127 put_nfs_open_dir_context(filp->private_data);
131 struct nfs_cache_array_entry {
135 unsigned char d_type;
138 struct nfs_cache_array {
142 struct nfs_cache_array_entry array[0];
145 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
149 unsigned long page_index;
152 loff_t current_index;
153 decode_dirent_t decode;
155 unsigned long timestamp;
156 unsigned long gencount;
157 unsigned int cache_entry_index;
160 } nfs_readdir_descriptor_t;
163 * The caller is responsible for calling nfs_readdir_release_array(page)
166 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
170 return ERR_PTR(-EIO);
173 return ERR_PTR(-ENOMEM);
178 void nfs_readdir_release_array(struct page *page)
184 * we are freeing strings created by nfs_add_to_readdir_array()
187 void nfs_readdir_clear_array(struct page *page)
189 struct nfs_cache_array *array;
192 array = kmap_atomic(page);
193 for (i = 0; i < array->size; i++)
194 kfree(array->array[i].string.name);
195 kunmap_atomic(array);
199 * the caller is responsible for freeing qstr.name
200 * when called by nfs_readdir_add_to_array, the strings will be freed in
201 * nfs_clear_readdir_array()
204 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
207 string->name = kmemdup(name, len, GFP_KERNEL);
208 if (string->name == NULL)
211 * Avoid a kmemleak false positive. The pointer to the name is stored
212 * in a page cache page which kmemleak does not scan.
214 kmemleak_not_leak(string->name);
215 string->hash = full_name_hash(name, len);
220 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
222 struct nfs_cache_array *array = nfs_readdir_get_array(page);
223 struct nfs_cache_array_entry *cache_entry;
227 return PTR_ERR(array);
229 cache_entry = &array->array[array->size];
231 /* Check that this entry lies within the page bounds */
233 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
236 cache_entry->cookie = entry->prev_cookie;
237 cache_entry->ino = entry->ino;
238 cache_entry->d_type = entry->d_type;
239 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
242 array->last_cookie = entry->cookie;
245 array->eof_index = array->size;
247 nfs_readdir_release_array(page);
252 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
254 loff_t diff = desc->file->f_pos - desc->current_index;
259 if (diff >= array->size) {
260 if (array->eof_index >= 0)
265 index = (unsigned int)diff;
266 *desc->dir_cookie = array->array[index].cookie;
267 desc->cache_entry_index = index;
275 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
279 int status = -EAGAIN;
281 for (i = 0; i < array->size; i++) {
282 if (array->array[i].cookie == *desc->dir_cookie) {
283 struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
284 struct nfs_open_dir_context *ctx = desc->file->private_data;
286 new_pos = desc->current_index + i;
287 if (ctx->attr_gencount != nfsi->attr_gencount
288 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
290 ctx->attr_gencount = nfsi->attr_gencount;
291 } else if (new_pos < desc->file->f_pos) {
293 && ctx->dup_cookie == *desc->dir_cookie) {
294 if (printk_ratelimit()) {
295 pr_notice("NFS: directory %s/%s contains a readdir loop."
296 "Please contact your server vendor. "
297 "The file: %s has duplicate cookie %llu\n",
298 desc->file->f_dentry->d_parent->d_name.name,
299 desc->file->f_dentry->d_name.name,
300 array->array[i].string.name,
306 ctx->dup_cookie = *desc->dir_cookie;
309 desc->file->f_pos = new_pos;
310 desc->cache_entry_index = i;
314 if (array->eof_index >= 0) {
315 status = -EBADCOOKIE;
316 if (*desc->dir_cookie == array->last_cookie)
324 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
326 struct nfs_cache_array *array;
329 array = nfs_readdir_get_array(desc->page);
331 status = PTR_ERR(array);
335 if (*desc->dir_cookie == 0)
336 status = nfs_readdir_search_for_pos(array, desc);
338 status = nfs_readdir_search_for_cookie(array, desc);
340 if (status == -EAGAIN) {
341 desc->last_cookie = array->last_cookie;
342 desc->current_index += array->size;
345 nfs_readdir_release_array(desc->page);
350 /* Fill a page with xdr information before transferring to the cache page */
352 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
353 struct nfs_entry *entry, struct file *file, struct inode *inode)
355 struct nfs_open_dir_context *ctx = file->private_data;
356 struct rpc_cred *cred = ctx->cred;
357 unsigned long timestamp, gencount;
362 gencount = nfs_inc_attr_generation_counter();
363 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
364 NFS_SERVER(inode)->dtsize, desc->plus);
366 /* We requested READDIRPLUS, but the server doesn't grok it */
367 if (error == -ENOTSUPP && desc->plus) {
368 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
369 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
375 desc->timestamp = timestamp;
376 desc->gencount = gencount;
381 static int xdr_decode(nfs_readdir_descriptor_t *desc,
382 struct nfs_entry *entry, struct xdr_stream *xdr)
386 error = desc->decode(xdr, entry, desc->plus);
389 entry->fattr->time_start = desc->timestamp;
390 entry->fattr->gencount = desc->gencount;
395 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
397 if (dentry->d_inode == NULL)
399 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
407 bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
409 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
411 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
413 if (filp->f_pos == 0)
419 * This function is called by the lookup code to request the use of
420 * readdirplus to accelerate any future lookups in the same
424 void nfs_advise_use_readdirplus(struct inode *dir)
426 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
430 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
432 struct qstr filename = QSTR_INIT(entry->name, entry->len);
433 struct dentry *dentry;
434 struct dentry *alias;
435 struct inode *dir = parent->d_inode;
438 if (filename.name[0] == '.') {
439 if (filename.len == 1)
441 if (filename.len == 2 && filename.name[1] == '.')
444 filename.hash = full_name_hash(filename.name, filename.len);
446 dentry = d_lookup(parent, &filename);
447 if (dentry != NULL) {
448 if (nfs_same_file(dentry, entry)) {
449 nfs_refresh_inode(dentry->d_inode, entry->fattr);
457 dentry = d_alloc(parent, &filename);
461 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
465 alias = d_materialise_unique(dentry, inode);
469 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
472 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
478 /* Perform conversion from xdr to cache array */
480 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
481 struct page **xdr_pages, struct page *page, unsigned int buflen)
483 struct xdr_stream stream;
485 struct page *scratch;
486 struct nfs_cache_array *array;
487 unsigned int count = 0;
490 scratch = alloc_page(GFP_KERNEL);
494 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
495 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
498 status = xdr_decode(desc, entry, &stream);
500 if (status == -EAGAIN)
508 nfs_prime_dcache(desc->file->f_path.dentry, entry);
510 status = nfs_readdir_add_to_array(entry, page);
513 } while (!entry->eof);
515 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
516 array = nfs_readdir_get_array(page);
517 if (!IS_ERR(array)) {
518 array->eof_index = array->size;
520 nfs_readdir_release_array(page);
522 status = PTR_ERR(array);
530 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
533 for (i = 0; i < npages; i++)
538 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
541 nfs_readdir_free_pagearray(pages, npages);
545 * nfs_readdir_large_page will allocate pages that must be freed with a call
546 * to nfs_readdir_free_large_page
549 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
553 for (i = 0; i < npages; i++) {
554 struct page *page = alloc_page(GFP_KERNEL);
562 nfs_readdir_free_pagearray(pages, i);
567 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
569 struct page *pages[NFS_MAX_READDIR_PAGES];
570 void *pages_ptr = NULL;
571 struct nfs_entry entry;
572 struct file *file = desc->file;
573 struct nfs_cache_array *array;
574 int status = -ENOMEM;
575 unsigned int array_size = ARRAY_SIZE(pages);
577 entry.prev_cookie = 0;
578 entry.cookie = desc->last_cookie;
580 entry.fh = nfs_alloc_fhandle();
581 entry.fattr = nfs_alloc_fattr();
582 entry.server = NFS_SERVER(inode);
583 if (entry.fh == NULL || entry.fattr == NULL)
586 array = nfs_readdir_get_array(page);
588 status = PTR_ERR(array);
591 memset(array, 0, sizeof(struct nfs_cache_array));
592 array->eof_index = -1;
594 status = nfs_readdir_large_page(pages, array_size);
596 goto out_release_array;
599 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
604 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
606 if (status == -ENOSPC)
610 } while (array->eof_index < 0);
612 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
614 nfs_readdir_release_array(page);
616 nfs_free_fattr(entry.fattr);
617 nfs_free_fhandle(entry.fh);
622 * Now we cache directories properly, by converting xdr information
623 * to an array that can be used for lookups later. This results in
624 * fewer cache pages, since we can store more information on each page.
625 * We only need to convert from xdr once so future lookups are much simpler
628 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
630 struct inode *inode = desc->file->f_path.dentry->d_inode;
633 ret = nfs_readdir_xdr_to_array(desc, page, inode);
636 SetPageUptodate(page);
638 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
639 /* Should never happen */
640 nfs_zap_mapping(inode, inode->i_mapping);
650 void cache_page_release(nfs_readdir_descriptor_t *desc)
652 if (!desc->page->mapping)
653 nfs_readdir_clear_array(desc->page);
654 page_cache_release(desc->page);
659 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
661 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
662 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
666 * Returns 0 if desc->dir_cookie was found on page desc->page_index
669 int find_cache_page(nfs_readdir_descriptor_t *desc)
673 desc->page = get_cache_page(desc);
674 if (IS_ERR(desc->page))
675 return PTR_ERR(desc->page);
677 res = nfs_readdir_search_array(desc);
679 cache_page_release(desc);
683 /* Search for desc->dir_cookie from the beginning of the page cache */
685 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
689 if (desc->page_index == 0) {
690 desc->current_index = 0;
691 desc->last_cookie = 0;
694 res = find_cache_page(desc);
695 } while (res == -EAGAIN);
700 * Once we've found the start of the dirent within a page: fill 'er up...
703 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
706 struct file *file = desc->file;
709 struct nfs_cache_array *array = NULL;
710 struct nfs_open_dir_context *ctx = file->private_data;
712 array = nfs_readdir_get_array(desc->page);
714 res = PTR_ERR(array);
718 for (i = desc->cache_entry_index; i < array->size; i++) {
719 struct nfs_cache_array_entry *ent;
721 ent = &array->array[i];
722 if (filldir(dirent, ent->string.name, ent->string.len,
723 file->f_pos, nfs_compat_user_ino64(ent->ino),
729 if (i < (array->size-1))
730 *desc->dir_cookie = array->array[i+1].cookie;
732 *desc->dir_cookie = array->last_cookie;
736 if (array->eof_index >= 0)
739 nfs_readdir_release_array(desc->page);
741 cache_page_release(desc);
742 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
743 (unsigned long long)*desc->dir_cookie, res);
748 * If we cannot find a cookie in our cache, we suspect that this is
749 * because it points to a deleted file, so we ask the server to return
750 * whatever it thinks is the next entry. We then feed this to filldir.
751 * If all goes well, we should then be able to find our way round the
752 * cache on the next call to readdir_search_pagecache();
754 * NOTE: we cannot add the anonymous page to the pagecache because
755 * the data it contains might not be page aligned. Besides,
756 * we should already have a complete representation of the
757 * directory in the page cache by the time we get here.
760 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
763 struct page *page = NULL;
765 struct inode *inode = desc->file->f_path.dentry->d_inode;
766 struct nfs_open_dir_context *ctx = desc->file->private_data;
768 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
769 (unsigned long long)*desc->dir_cookie);
771 page = alloc_page(GFP_HIGHUSER);
777 desc->page_index = 0;
778 desc->last_cookie = *desc->dir_cookie;
782 status = nfs_readdir_xdr_to_array(desc, page, inode);
786 status = nfs_do_filldir(desc, dirent, filldir);
789 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
793 cache_page_release(desc);
797 /* The file offset position represents the dirent entry number. A
798 last cookie cache takes care of the common case of reading the
801 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
803 struct dentry *dentry = filp->f_path.dentry;
804 struct inode *inode = dentry->d_inode;
805 nfs_readdir_descriptor_t my_desc,
807 struct nfs_open_dir_context *dir_ctx = filp->private_data;
810 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
811 dentry->d_parent->d_name.name, dentry->d_name.name,
812 (long long)filp->f_pos);
813 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
816 * filp->f_pos points to the dirent entry number.
817 * *desc->dir_cookie has the cookie for the next entry. We have
818 * to either find the entry with the appropriate number or
819 * revalidate the cookie.
821 memset(desc, 0, sizeof(*desc));
824 desc->dir_cookie = &dir_ctx->dir_cookie;
825 desc->decode = NFS_PROTO(inode)->decode_dirent;
826 desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
828 nfs_block_sillyrename(dentry);
829 res = nfs_revalidate_mapping(inode, filp->f_mapping);
834 res = readdir_search_pagecache(desc);
836 if (res == -EBADCOOKIE) {
838 /* This means either end of directory */
839 if (*desc->dir_cookie && desc->eof == 0) {
840 /* Or that the server has 'lost' a cookie */
841 res = uncached_readdir(desc, dirent, filldir);
847 if (res == -ETOOSMALL && desc->plus) {
848 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
849 nfs_zap_caches(inode);
850 desc->page_index = 0;
858 res = nfs_do_filldir(desc, dirent, filldir);
861 } while (!desc->eof);
863 nfs_unblock_sillyrename(dentry);
866 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
867 dentry->d_parent->d_name.name, dentry->d_name.name,
872 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
874 struct dentry *dentry = filp->f_path.dentry;
875 struct inode *inode = dentry->d_inode;
876 struct nfs_open_dir_context *dir_ctx = filp->private_data;
878 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
879 dentry->d_parent->d_name.name,
883 mutex_lock(&inode->i_mutex);
886 offset += filp->f_pos;
894 if (offset != filp->f_pos) {
895 filp->f_pos = offset;
896 dir_ctx->dir_cookie = 0;
900 mutex_unlock(&inode->i_mutex);
905 * All directory operations under NFS are synchronous, so fsync()
906 * is a dummy operation.
908 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
911 struct dentry *dentry = filp->f_path.dentry;
912 struct inode *inode = dentry->d_inode;
914 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
915 dentry->d_parent->d_name.name, dentry->d_name.name,
918 mutex_lock(&inode->i_mutex);
919 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
920 mutex_unlock(&inode->i_mutex);
925 * nfs_force_lookup_revalidate - Mark the directory as having changed
926 * @dir - pointer to directory inode
928 * This forces the revalidation code in nfs_lookup_revalidate() to do a
929 * full lookup on all child dentries of 'dir' whenever a change occurs
930 * on the server that might have invalidated our dcache.
932 * The caller should be holding dir->i_lock
934 void nfs_force_lookup_revalidate(struct inode *dir)
936 NFS_I(dir)->cache_change_attribute++;
940 * A check for whether or not the parent directory has changed.
941 * In the case it has, we assume that the dentries are untrustworthy
942 * and may need to be looked up again.
944 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
948 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
950 if (!nfs_verify_change_attribute(dir, dentry->d_time))
952 /* Revalidate nfsi->cache_change_attribute before we declare a match */
953 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
955 if (!nfs_verify_change_attribute(dir, dentry->d_time))
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, unsigned int flags)
966 if (NFS_PROTO(dir)->version == 2)
968 return flags & 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, unsigned int flags)
982 struct nfs_server *server = NFS_SERVER(inode);
984 if (IS_AUTOMOUNT(inode))
986 /* VFS wants an on-the-wire revalidation */
987 if (flags & LOOKUP_REVAL)
989 /* This is an open(2) */
990 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
991 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
995 return __nfs_revalidate_inode(server, inode);
999 * We judge how long we want to trust negative
1000 * dentries by looking at the parent inode mtime.
1002 * If parent mtime has changed, we revalidate, else we wait for a
1003 * period corresponding to the parent's attribute cache timeout value.
1006 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1009 /* Don't revalidate a negative dentry if we're creating a new file */
1010 if (flags & LOOKUP_CREATE)
1012 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1014 return !nfs_check_verifier(dir, dentry);
1018 * This is called every time the dcache has a lookup hit,
1019 * and we should check whether we can really trust that
1022 * NOTE! The hit can be a negative hit too, don't assume
1025 * If the parent directory is seen to have changed, we throw out the
1026 * cached dentry and do a new lookup.
1028 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1031 struct inode *inode;
1032 struct dentry *parent;
1033 struct nfs_fh *fhandle = NULL;
1034 struct nfs_fattr *fattr = NULL;
1037 if (flags & LOOKUP_RCU)
1040 parent = dget_parent(dentry);
1041 dir = parent->d_inode;
1042 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1043 inode = dentry->d_inode;
1046 if (nfs_neg_need_reval(dir, dentry, flags))
1048 goto out_valid_noent;
1051 if (is_bad_inode(inode)) {
1052 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1053 __func__, dentry->d_parent->d_name.name,
1054 dentry->d_name.name);
1058 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1059 goto out_set_verifier;
1061 /* Force a full look up iff the parent directory has changed */
1062 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1063 if (nfs_lookup_verify_inode(inode, flags))
1064 goto out_zap_parent;
1068 if (NFS_STALE(inode))
1072 fhandle = nfs_alloc_fhandle();
1073 fattr = nfs_alloc_fattr();
1074 if (fhandle == NULL || fattr == NULL)
1077 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1080 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1082 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1085 nfs_free_fattr(fattr);
1086 nfs_free_fhandle(fhandle);
1088 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1090 /* Success: notify readdir to use READDIRPLUS */
1091 nfs_advise_use_readdirplus(dir);
1094 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1095 __func__, dentry->d_parent->d_name.name,
1096 dentry->d_name.name);
1099 nfs_zap_caches(dir);
1101 nfs_mark_for_revalidate(dir);
1102 if (inode && S_ISDIR(inode->i_mode)) {
1103 /* Purge readdir caches. */
1104 nfs_zap_caches(inode);
1105 /* If we have submounts, don't unhash ! */
1106 if (have_submounts(dentry))
1108 if (dentry->d_flags & DCACHE_DISCONNECTED)
1110 shrink_dcache_parent(dentry);
1113 nfs_free_fattr(fattr);
1114 nfs_free_fhandle(fhandle);
1116 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1117 __func__, dentry->d_parent->d_name.name,
1118 dentry->d_name.name);
1121 nfs_free_fattr(fattr);
1122 nfs_free_fhandle(fhandle);
1124 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1125 __func__, dentry->d_parent->d_name.name,
1126 dentry->d_name.name, error);
1131 * This is called from dput() when d_count is going to 0.
1133 static int nfs_dentry_delete(const struct dentry *dentry)
1135 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1136 dentry->d_parent->d_name.name, dentry->d_name.name,
1139 /* Unhash any dentry with a stale inode */
1140 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1143 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1144 /* Unhash it, so that ->d_iput() would be called */
1147 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1148 /* Unhash it, so that ancestors of killed async unlink
1149 * files will be cleaned up during umount */
1156 static void nfs_drop_nlink(struct inode *inode)
1158 spin_lock(&inode->i_lock);
1159 if (inode->i_nlink > 0)
1161 spin_unlock(&inode->i_lock);
1165 * Called when the dentry loses inode.
1166 * We use it to clean up silly-renamed files.
1168 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1170 if (S_ISDIR(inode->i_mode))
1171 /* drop any readdir cache as it could easily be old */
1172 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1174 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1176 nfs_complete_unlink(dentry, inode);
1181 static void nfs_d_release(struct dentry *dentry)
1183 /* free cached devname value, if it survived that far */
1184 if (unlikely(dentry->d_fsdata)) {
1185 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1188 kfree(dentry->d_fsdata);
1192 const struct dentry_operations nfs_dentry_operations = {
1193 .d_revalidate = nfs_lookup_revalidate,
1194 .d_delete = nfs_dentry_delete,
1195 .d_iput = nfs_dentry_iput,
1196 .d_automount = nfs_d_automount,
1197 .d_release = nfs_d_release,
1200 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1203 struct dentry *parent;
1204 struct inode *inode = NULL;
1205 struct nfs_fh *fhandle = NULL;
1206 struct nfs_fattr *fattr = NULL;
1209 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1210 dentry->d_parent->d_name.name, dentry->d_name.name);
1211 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1213 res = ERR_PTR(-ENAMETOOLONG);
1214 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1218 * If we're doing an exclusive create, optimize away the lookup
1219 * but don't hash the dentry.
1221 if (nfs_is_exclusive_create(dir, flags)) {
1222 d_instantiate(dentry, NULL);
1227 res = ERR_PTR(-ENOMEM);
1228 fhandle = nfs_alloc_fhandle();
1229 fattr = nfs_alloc_fattr();
1230 if (fhandle == NULL || fattr == NULL)
1233 parent = dentry->d_parent;
1234 /* Protect against concurrent sillydeletes */
1235 nfs_block_sillyrename(parent);
1236 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1237 if (error == -ENOENT)
1240 res = ERR_PTR(error);
1241 goto out_unblock_sillyrename;
1243 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1244 res = ERR_CAST(inode);
1246 goto out_unblock_sillyrename;
1248 /* Success: notify readdir to use READDIRPLUS */
1249 nfs_advise_use_readdirplus(dir);
1252 res = d_materialise_unique(dentry, inode);
1255 goto out_unblock_sillyrename;
1258 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1259 out_unblock_sillyrename:
1260 nfs_unblock_sillyrename(parent);
1262 nfs_free_fattr(fattr);
1263 nfs_free_fhandle(fhandle);
1267 #ifdef CONFIG_NFS_V4
1268 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1270 const struct dentry_operations nfs4_dentry_operations = {
1271 .d_revalidate = nfs4_lookup_revalidate,
1272 .d_delete = nfs_dentry_delete,
1273 .d_iput = nfs_dentry_iput,
1274 .d_automount = nfs_d_automount,
1275 .d_release = nfs_d_release,
1278 static fmode_t flags_to_mode(int flags)
1280 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1281 if ((flags & O_ACCMODE) != O_WRONLY)
1283 if ((flags & O_ACCMODE) != O_RDONLY)
1288 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1290 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1293 static int do_open(struct inode *inode, struct file *filp)
1295 nfs_fscache_set_inode_cookie(inode, filp);
1299 static int nfs_finish_open(struct nfs_open_context *ctx,
1300 struct dentry *dentry,
1301 struct file *file, unsigned open_flags,
1306 if (ctx->dentry != dentry) {
1308 ctx->dentry = dget(dentry);
1311 /* If the open_intent is for execute, we have an extra check to make */
1312 if (ctx->mode & FMODE_EXEC) {
1313 err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1318 err = finish_open(file, dentry, do_open, opened);
1321 nfs_file_set_open_context(file, ctx);
1324 put_nfs_open_context(ctx);
1328 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1329 struct file *file, unsigned open_flags,
1330 umode_t mode, int *opened)
1332 struct nfs_open_context *ctx;
1334 struct iattr attr = { .ia_valid = ATTR_OPEN };
1335 struct inode *inode;
1338 /* Expect a negative dentry */
1339 BUG_ON(dentry->d_inode);
1341 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1342 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1344 /* NFS only supports OPEN on regular files */
1345 if ((open_flags & O_DIRECTORY)) {
1346 if (!d_unhashed(dentry)) {
1348 * Hashed negative dentry with O_DIRECTORY: dentry was
1349 * revalidated and is fine, no need to perform lookup
1357 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1358 return -ENAMETOOLONG;
1360 if (open_flags & O_CREAT) {
1361 attr.ia_valid |= ATTR_MODE;
1362 attr.ia_mode = mode & ~current_umask();
1364 if (open_flags & O_TRUNC) {
1365 attr.ia_valid |= ATTR_SIZE;
1369 ctx = create_nfs_open_context(dentry, open_flags);
1374 nfs_block_sillyrename(dentry->d_parent);
1375 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1377 if (IS_ERR(inode)) {
1378 nfs_unblock_sillyrename(dentry->d_parent);
1379 put_nfs_open_context(ctx);
1380 err = PTR_ERR(inode);
1383 d_add(dentry, NULL);
1389 if (!(open_flags & O_NOFOLLOW))
1398 res = d_add_unique(dentry, inode);
1402 nfs_unblock_sillyrename(dentry->d_parent);
1403 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1405 err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1412 res = nfs_lookup(dir, dentry, 0);
1417 return finish_no_open(file, res);
1420 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1422 struct dentry *parent = NULL;
1423 struct inode *inode;
1427 if (flags & LOOKUP_RCU)
1430 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1432 if (d_mountpoint(dentry))
1435 inode = dentry->d_inode;
1436 parent = dget_parent(dentry);
1437 dir = parent->d_inode;
1439 /* We can't create new files in nfs_open_revalidate(), so we
1440 * optimize away revalidation of negative dentries.
1442 if (inode == NULL) {
1443 if (!nfs_neg_need_reval(dir, dentry, flags))
1448 /* NFS only supports OPEN on regular files */
1449 if (!S_ISREG(inode->i_mode))
1451 /* We cannot do exclusive creation on a positive dentry */
1452 if (flags & LOOKUP_EXCL)
1455 /* Let f_op->open() actually open (and revalidate) the file */
1465 return nfs_lookup_revalidate(dentry, flags);
1468 #endif /* CONFIG_NFSV4 */
1471 * Code common to create, mkdir, and mknod.
1473 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1474 struct nfs_fattr *fattr)
1476 struct dentry *parent = dget_parent(dentry);
1477 struct inode *dir = parent->d_inode;
1478 struct inode *inode;
1479 int error = -EACCES;
1483 /* We may have been initialized further down */
1484 if (dentry->d_inode)
1486 if (fhandle->size == 0) {
1487 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1491 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1492 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1493 struct nfs_server *server = NFS_SB(dentry->d_sb);
1494 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1498 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1499 error = PTR_ERR(inode);
1502 d_add(dentry, inode);
1507 nfs_mark_for_revalidate(dir);
1513 * Following a failed create operation, we drop the dentry rather
1514 * than retain a negative dentry. This avoids a problem in the event
1515 * that the operation succeeded on the server, but an error in the
1516 * reply path made it appear to have failed.
1518 int nfs_create(struct inode *dir, struct dentry *dentry,
1519 umode_t mode, bool excl)
1522 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1525 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1526 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1528 attr.ia_mode = mode;
1529 attr.ia_valid = ATTR_MODE;
1531 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1541 * See comments for nfs_proc_create regarding failed operations.
1544 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1549 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1550 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1552 if (!new_valid_dev(rdev))
1555 attr.ia_mode = mode;
1556 attr.ia_valid = ATTR_MODE;
1558 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1568 * See comments for nfs_proc_create regarding failed operations.
1570 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1575 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1576 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1578 attr.ia_valid = ATTR_MODE;
1579 attr.ia_mode = mode | S_IFDIR;
1581 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1590 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1592 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1596 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1600 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1601 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1603 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1604 /* Ensure the VFS deletes this inode */
1605 if (error == 0 && dentry->d_inode != NULL)
1606 clear_nlink(dentry->d_inode);
1607 else if (error == -ENOENT)
1608 nfs_dentry_handle_enoent(dentry);
1614 * Remove a file after making sure there are no pending writes,
1615 * and after checking that the file has only one user.
1617 * We invalidate the attribute cache and free the inode prior to the operation
1618 * to avoid possible races if the server reuses the inode.
1620 static int nfs_safe_remove(struct dentry *dentry)
1622 struct inode *dir = dentry->d_parent->d_inode;
1623 struct inode *inode = dentry->d_inode;
1626 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1627 dentry->d_parent->d_name.name, dentry->d_name.name);
1629 /* If the dentry was sillyrenamed, we simply call d_delete() */
1630 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1635 if (inode != NULL) {
1636 NFS_PROTO(inode)->return_delegation(inode);
1637 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1638 /* The VFS may want to delete this inode */
1640 nfs_drop_nlink(inode);
1641 nfs_mark_for_revalidate(inode);
1643 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1644 if (error == -ENOENT)
1645 nfs_dentry_handle_enoent(dentry);
1650 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1651 * belongs to an active ".nfs..." file and we return -EBUSY.
1653 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1655 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1658 int need_rehash = 0;
1660 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1661 dir->i_ino, dentry->d_name.name);
1663 spin_lock(&dentry->d_lock);
1664 if (dentry->d_count > 1) {
1665 spin_unlock(&dentry->d_lock);
1666 /* Start asynchronous writeout of the inode */
1667 write_inode_now(dentry->d_inode, 0);
1668 error = nfs_sillyrename(dir, dentry);
1671 if (!d_unhashed(dentry)) {
1675 spin_unlock(&dentry->d_lock);
1676 error = nfs_safe_remove(dentry);
1677 if (!error || error == -ENOENT) {
1678 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1679 } else if (need_rehash)
1685 * To create a symbolic link, most file systems instantiate a new inode,
1686 * add a page to it containing the path, then write it out to the disk
1687 * using prepare_write/commit_write.
1689 * Unfortunately the NFS client can't create the in-core inode first
1690 * because it needs a file handle to create an in-core inode (see
1691 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1692 * symlink request has completed on the server.
1694 * So instead we allocate a raw page, copy the symname into it, then do
1695 * the SYMLINK request with the page as the buffer. If it succeeds, we
1696 * now have a new file handle and can instantiate an in-core NFS inode
1697 * and move the raw page into its mapping.
1699 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1701 struct pagevec lru_pvec;
1705 unsigned int pathlen = strlen(symname);
1708 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1709 dir->i_ino, dentry->d_name.name, symname);
1711 if (pathlen > PAGE_SIZE)
1712 return -ENAMETOOLONG;
1714 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1715 attr.ia_valid = ATTR_MODE;
1717 page = alloc_page(GFP_HIGHUSER);
1721 kaddr = kmap_atomic(page);
1722 memcpy(kaddr, symname, pathlen);
1723 if (pathlen < PAGE_SIZE)
1724 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1725 kunmap_atomic(kaddr);
1727 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1729 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1730 dir->i_sb->s_id, dir->i_ino,
1731 dentry->d_name.name, symname, error);
1738 * No big deal if we can't add this page to the page cache here.
1739 * READLINK will get the missing page from the server if needed.
1741 pagevec_init(&lru_pvec, 0);
1742 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1744 pagevec_add(&lru_pvec, page);
1745 pagevec_lru_add_file(&lru_pvec);
1746 SetPageUptodate(page);
1755 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1757 struct inode *inode = old_dentry->d_inode;
1760 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1761 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1762 dentry->d_parent->d_name.name, dentry->d_name.name);
1764 NFS_PROTO(inode)->return_delegation(inode);
1767 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1770 d_add(dentry, inode);
1777 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1778 * different file handle for the same inode after a rename (e.g. when
1779 * moving to a different directory). A fail-safe method to do so would
1780 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1781 * rename the old file using the sillyrename stuff. This way, the original
1782 * file in old_dir will go away when the last process iput()s the inode.
1786 * It actually works quite well. One needs to have the possibility for
1787 * at least one ".nfs..." file in each directory the file ever gets
1788 * moved or linked to which happens automagically with the new
1789 * implementation that only depends on the dcache stuff instead of
1790 * using the inode layer
1792 * Unfortunately, things are a little more complicated than indicated
1793 * above. For a cross-directory move, we want to make sure we can get
1794 * rid of the old inode after the operation. This means there must be
1795 * no pending writes (if it's a file), and the use count must be 1.
1796 * If these conditions are met, we can drop the dentries before doing
1799 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1800 struct inode *new_dir, struct dentry *new_dentry)
1802 struct inode *old_inode = old_dentry->d_inode;
1803 struct inode *new_inode = new_dentry->d_inode;
1804 struct dentry *dentry = NULL, *rehash = NULL;
1807 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1808 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1809 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1810 new_dentry->d_count);
1813 * For non-directories, check whether the target is busy and if so,
1814 * make a copy of the dentry and then do a silly-rename. If the
1815 * silly-rename succeeds, the copied dentry is hashed and becomes
1818 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1820 * To prevent any new references to the target during the
1821 * rename, we unhash the dentry in advance.
1823 if (!d_unhashed(new_dentry)) {
1825 rehash = new_dentry;
1828 if (new_dentry->d_count > 2) {
1831 /* copy the target dentry's name */
1832 dentry = d_alloc(new_dentry->d_parent,
1833 &new_dentry->d_name);
1837 /* silly-rename the existing target ... */
1838 err = nfs_sillyrename(new_dir, new_dentry);
1842 new_dentry = dentry;
1848 NFS_PROTO(old_inode)->return_delegation(old_inode);
1849 if (new_inode != NULL)
1850 NFS_PROTO(new_inode)->return_delegation(new_inode);
1852 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1853 new_dir, &new_dentry->d_name);
1854 nfs_mark_for_revalidate(old_inode);
1859 if (new_inode != NULL)
1860 nfs_drop_nlink(new_inode);
1861 d_move(old_dentry, new_dentry);
1862 nfs_set_verifier(new_dentry,
1863 nfs_save_change_attribute(new_dir));
1864 } else if (error == -ENOENT)
1865 nfs_dentry_handle_enoent(old_dentry);
1867 /* new dentry created? */
1873 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1874 static LIST_HEAD(nfs_access_lru_list);
1875 static atomic_long_t nfs_access_nr_entries;
1877 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1879 put_rpccred(entry->cred);
1881 smp_mb__before_atomic_dec();
1882 atomic_long_dec(&nfs_access_nr_entries);
1883 smp_mb__after_atomic_dec();
1886 static void nfs_access_free_list(struct list_head *head)
1888 struct nfs_access_entry *cache;
1890 while (!list_empty(head)) {
1891 cache = list_entry(head->next, struct nfs_access_entry, lru);
1892 list_del(&cache->lru);
1893 nfs_access_free_entry(cache);
1897 int nfs_access_cache_shrinker(struct shrinker *shrink,
1898 struct shrink_control *sc)
1901 struct nfs_inode *nfsi, *next;
1902 struct nfs_access_entry *cache;
1903 int nr_to_scan = sc->nr_to_scan;
1904 gfp_t gfp_mask = sc->gfp_mask;
1906 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1907 return (nr_to_scan == 0) ? 0 : -1;
1909 spin_lock(&nfs_access_lru_lock);
1910 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1911 struct inode *inode;
1913 if (nr_to_scan-- == 0)
1915 inode = &nfsi->vfs_inode;
1916 spin_lock(&inode->i_lock);
1917 if (list_empty(&nfsi->access_cache_entry_lru))
1918 goto remove_lru_entry;
1919 cache = list_entry(nfsi->access_cache_entry_lru.next,
1920 struct nfs_access_entry, lru);
1921 list_move(&cache->lru, &head);
1922 rb_erase(&cache->rb_node, &nfsi->access_cache);
1923 if (!list_empty(&nfsi->access_cache_entry_lru))
1924 list_move_tail(&nfsi->access_cache_inode_lru,
1925 &nfs_access_lru_list);
1928 list_del_init(&nfsi->access_cache_inode_lru);
1929 smp_mb__before_clear_bit();
1930 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1931 smp_mb__after_clear_bit();
1933 spin_unlock(&inode->i_lock);
1935 spin_unlock(&nfs_access_lru_lock);
1936 nfs_access_free_list(&head);
1937 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1940 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1942 struct rb_root *root_node = &nfsi->access_cache;
1944 struct nfs_access_entry *entry;
1946 /* Unhook entries from the cache */
1947 while ((n = rb_first(root_node)) != NULL) {
1948 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1949 rb_erase(n, root_node);
1950 list_move(&entry->lru, head);
1952 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1955 void nfs_access_zap_cache(struct inode *inode)
1959 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
1961 /* Remove from global LRU init */
1962 spin_lock(&nfs_access_lru_lock);
1963 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
1964 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1966 spin_lock(&inode->i_lock);
1967 __nfs_access_zap_cache(NFS_I(inode), &head);
1968 spin_unlock(&inode->i_lock);
1969 spin_unlock(&nfs_access_lru_lock);
1970 nfs_access_free_list(&head);
1973 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1975 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1976 struct nfs_access_entry *entry;
1979 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1981 if (cred < entry->cred)
1983 else if (cred > entry->cred)
1991 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1993 struct nfs_inode *nfsi = NFS_I(inode);
1994 struct nfs_access_entry *cache;
1997 spin_lock(&inode->i_lock);
1998 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2000 cache = nfs_access_search_rbtree(inode, cred);
2003 if (!nfs_have_delegated_attributes(inode) &&
2004 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2006 res->jiffies = cache->jiffies;
2007 res->cred = cache->cred;
2008 res->mask = cache->mask;
2009 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2012 spin_unlock(&inode->i_lock);
2015 rb_erase(&cache->rb_node, &nfsi->access_cache);
2016 list_del(&cache->lru);
2017 spin_unlock(&inode->i_lock);
2018 nfs_access_free_entry(cache);
2021 spin_unlock(&inode->i_lock);
2022 nfs_access_zap_cache(inode);
2026 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2028 struct nfs_inode *nfsi = NFS_I(inode);
2029 struct rb_root *root_node = &nfsi->access_cache;
2030 struct rb_node **p = &root_node->rb_node;
2031 struct rb_node *parent = NULL;
2032 struct nfs_access_entry *entry;
2034 spin_lock(&inode->i_lock);
2035 while (*p != NULL) {
2037 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2039 if (set->cred < entry->cred)
2040 p = &parent->rb_left;
2041 else if (set->cred > entry->cred)
2042 p = &parent->rb_right;
2046 rb_link_node(&set->rb_node, parent, p);
2047 rb_insert_color(&set->rb_node, root_node);
2048 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2049 spin_unlock(&inode->i_lock);
2052 rb_replace_node(parent, &set->rb_node, root_node);
2053 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2054 list_del(&entry->lru);
2055 spin_unlock(&inode->i_lock);
2056 nfs_access_free_entry(entry);
2059 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2061 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2064 RB_CLEAR_NODE(&cache->rb_node);
2065 cache->jiffies = set->jiffies;
2066 cache->cred = get_rpccred(set->cred);
2067 cache->mask = set->mask;
2069 nfs_access_add_rbtree(inode, cache);
2071 /* Update accounting */
2072 smp_mb__before_atomic_inc();
2073 atomic_long_inc(&nfs_access_nr_entries);
2074 smp_mb__after_atomic_inc();
2076 /* Add inode to global LRU list */
2077 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2078 spin_lock(&nfs_access_lru_lock);
2079 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2080 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2081 &nfs_access_lru_list);
2082 spin_unlock(&nfs_access_lru_lock);
2086 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2088 struct nfs_access_entry cache;
2091 status = nfs_access_get_cached(inode, cred, &cache);
2095 /* Be clever: ask server to check for all possible rights */
2096 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2098 cache.jiffies = jiffies;
2099 status = NFS_PROTO(inode)->access(inode, &cache);
2101 if (status == -ESTALE) {
2102 nfs_zap_caches(inode);
2103 if (!S_ISDIR(inode->i_mode))
2104 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2108 nfs_access_add_cache(inode, &cache);
2110 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2115 static int nfs_open_permission_mask(int openflags)
2119 if ((openflags & O_ACCMODE) != O_WRONLY)
2121 if ((openflags & O_ACCMODE) != O_RDONLY)
2123 if (openflags & __FMODE_EXEC)
2128 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2130 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2133 int nfs_permission(struct inode *inode, int mask)
2135 struct rpc_cred *cred;
2138 if (mask & MAY_NOT_BLOCK)
2141 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2143 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2145 /* Is this sys_access() ? */
2146 if (mask & (MAY_ACCESS | MAY_CHDIR))
2149 switch (inode->i_mode & S_IFMT) {
2153 /* NFSv4 has atomic_open... */
2154 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2155 && (mask & MAY_OPEN)
2156 && !(mask & MAY_EXEC))
2161 * Optimize away all write operations, since the server
2162 * will check permissions when we perform the op.
2164 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2169 if (!NFS_PROTO(inode)->access)
2172 cred = rpc_lookup_cred();
2173 if (!IS_ERR(cred)) {
2174 res = nfs_do_access(inode, cred, mask);
2177 res = PTR_ERR(cred);
2179 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2182 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2183 inode->i_sb->s_id, inode->i_ino, mask, res);
2186 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2188 res = generic_permission(inode, mask);
2194 * version-control: t
2195 * kept-new-versions: 5