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/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/sched.h>
37 #include <linux/kmemleak.h>
38 #include <linux/xattr.h>
40 #include "delegation.h"
45 /* #define NFS_DEBUG_VERBOSE 1 */
47 static int nfs_opendir(struct inode *, struct file *);
48 static int nfs_closedir(struct inode *, struct file *);
49 static int nfs_readdir(struct file *, struct dir_context *);
50 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
51 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
52 static void nfs_readdir_clear_array(struct page*);
54 const struct file_operations nfs_dir_operations = {
55 .llseek = nfs_llseek_dir,
56 .read = generic_read_dir,
57 .iterate = nfs_readdir,
59 .release = nfs_closedir,
60 .fsync = nfs_fsync_dir,
63 const struct address_space_operations nfs_dir_aops = {
64 .freepage = nfs_readdir_clear_array,
67 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
69 struct nfs_open_dir_context *ctx;
70 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
73 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
76 ctx->cred = get_rpccred(cred);
79 return ERR_PTR(-ENOMEM);
82 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
84 put_rpccred(ctx->cred);
92 nfs_opendir(struct inode *inode, struct file *filp)
95 struct nfs_open_dir_context *ctx;
96 struct rpc_cred *cred;
98 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
99 filp->f_path.dentry->d_parent->d_name.name,
100 filp->f_path.dentry->d_name.name);
102 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
104 cred = rpc_lookup_cred();
106 return PTR_ERR(cred);
107 ctx = alloc_nfs_open_dir_context(inode, cred);
112 filp->private_data = ctx;
113 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
114 /* This is a mountpoint, so d_revalidate will never
115 * have been called, so we need to refresh the
116 * inode (for close-open consistency) ourselves.
118 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
126 nfs_closedir(struct inode *inode, struct file *filp)
128 put_nfs_open_dir_context(filp->private_data);
132 struct nfs_cache_array_entry {
136 unsigned char d_type;
139 struct nfs_cache_array {
143 struct nfs_cache_array_entry array[0];
146 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
150 struct dir_context *ctx;
151 unsigned long page_index;
154 loff_t current_index;
155 decode_dirent_t decode;
157 unsigned long timestamp;
158 unsigned long gencount;
159 unsigned int cache_entry_index;
162 } nfs_readdir_descriptor_t;
165 * The caller is responsible for calling nfs_readdir_release_array(page)
168 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
172 return ERR_PTR(-EIO);
175 return ERR_PTR(-ENOMEM);
180 void nfs_readdir_release_array(struct page *page)
186 * we are freeing strings created by nfs_add_to_readdir_array()
189 void nfs_readdir_clear_array(struct page *page)
191 struct nfs_cache_array *array;
194 array = kmap_atomic(page);
195 for (i = 0; i < array->size; i++)
196 kfree(array->array[i].string.name);
197 kunmap_atomic(array);
201 * the caller is responsible for freeing qstr.name
202 * when called by nfs_readdir_add_to_array, the strings will be freed in
203 * nfs_clear_readdir_array()
206 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
209 string->name = kmemdup(name, len, GFP_KERNEL);
210 if (string->name == NULL)
213 * Avoid a kmemleak false positive. The pointer to the name is stored
214 * in a page cache page which kmemleak does not scan.
216 kmemleak_not_leak(string->name);
217 string->hash = full_name_hash(name, len);
222 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
224 struct nfs_cache_array *array = nfs_readdir_get_array(page);
225 struct nfs_cache_array_entry *cache_entry;
229 return PTR_ERR(array);
231 cache_entry = &array->array[array->size];
233 /* Check that this entry lies within the page bounds */
235 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
238 cache_entry->cookie = entry->prev_cookie;
239 cache_entry->ino = entry->ino;
240 cache_entry->d_type = entry->d_type;
241 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
244 array->last_cookie = entry->cookie;
247 array->eof_index = array->size;
249 nfs_readdir_release_array(page);
254 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
256 loff_t diff = desc->ctx->pos - desc->current_index;
261 if (diff >= array->size) {
262 if (array->eof_index >= 0)
267 index = (unsigned int)diff;
268 *desc->dir_cookie = array->array[index].cookie;
269 desc->cache_entry_index = index;
277 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
281 int status = -EAGAIN;
283 for (i = 0; i < array->size; i++) {
284 if (array->array[i].cookie == *desc->dir_cookie) {
285 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
286 struct nfs_open_dir_context *ctx = desc->file->private_data;
288 new_pos = desc->current_index + i;
289 if (ctx->attr_gencount != nfsi->attr_gencount
290 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
292 ctx->attr_gencount = nfsi->attr_gencount;
293 } else if (new_pos < desc->ctx->pos) {
295 && ctx->dup_cookie == *desc->dir_cookie) {
296 if (printk_ratelimit()) {
297 pr_notice("NFS: directory %s/%s contains a readdir loop."
298 "Please contact your server vendor. "
299 "The file: %s has duplicate cookie %llu\n",
300 desc->file->f_dentry->d_parent->d_name.name,
301 desc->file->f_dentry->d_name.name,
302 array->array[i].string.name,
308 ctx->dup_cookie = *desc->dir_cookie;
311 desc->ctx->pos = new_pos;
312 desc->cache_entry_index = i;
316 if (array->eof_index >= 0) {
317 status = -EBADCOOKIE;
318 if (*desc->dir_cookie == array->last_cookie)
326 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
328 struct nfs_cache_array *array;
331 array = nfs_readdir_get_array(desc->page);
333 status = PTR_ERR(array);
337 if (*desc->dir_cookie == 0)
338 status = nfs_readdir_search_for_pos(array, desc);
340 status = nfs_readdir_search_for_cookie(array, desc);
342 if (status == -EAGAIN) {
343 desc->last_cookie = array->last_cookie;
344 desc->current_index += array->size;
347 nfs_readdir_release_array(desc->page);
352 /* Fill a page with xdr information before transferring to the cache page */
354 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
355 struct nfs_entry *entry, struct file *file, struct inode *inode)
357 struct nfs_open_dir_context *ctx = file->private_data;
358 struct rpc_cred *cred = ctx->cred;
359 unsigned long timestamp, gencount;
364 gencount = nfs_inc_attr_generation_counter();
365 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
366 NFS_SERVER(inode)->dtsize, desc->plus);
368 /* We requested READDIRPLUS, but the server doesn't grok it */
369 if (error == -ENOTSUPP && desc->plus) {
370 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
371 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
377 desc->timestamp = timestamp;
378 desc->gencount = gencount;
383 static int xdr_decode(nfs_readdir_descriptor_t *desc,
384 struct nfs_entry *entry, struct xdr_stream *xdr)
388 error = desc->decode(xdr, entry, desc->plus);
391 entry->fattr->time_start = desc->timestamp;
392 entry->fattr->gencount = desc->gencount;
397 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
399 if (dentry->d_inode == NULL)
401 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
409 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
411 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
413 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
421 * This function is called by the lookup code to request the use of
422 * readdirplus to accelerate any future lookups in the same
426 void nfs_advise_use_readdirplus(struct inode *dir)
428 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
432 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
434 struct qstr filename = QSTR_INIT(entry->name, entry->len);
435 struct dentry *dentry;
436 struct dentry *alias;
437 struct inode *dir = parent->d_inode;
440 if (filename.name[0] == '.') {
441 if (filename.len == 1)
443 if (filename.len == 2 && filename.name[1] == '.')
446 filename.hash = full_name_hash(filename.name, filename.len);
448 dentry = d_lookup(parent, &filename);
449 if (dentry != NULL) {
450 if (nfs_same_file(dentry, entry)) {
451 nfs_refresh_inode(dentry->d_inode, entry->fattr);
454 if (d_invalidate(dentry) != 0)
460 dentry = d_alloc(parent, &filename);
464 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
468 alias = d_materialise_unique(dentry, inode);
472 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
475 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
481 /* Perform conversion from xdr to cache array */
483 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
484 struct page **xdr_pages, struct page *page, unsigned int buflen)
486 struct xdr_stream stream;
488 struct page *scratch;
489 struct nfs_cache_array *array;
490 unsigned int count = 0;
493 scratch = alloc_page(GFP_KERNEL);
497 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
498 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
501 status = xdr_decode(desc, entry, &stream);
503 if (status == -EAGAIN)
511 nfs_prime_dcache(desc->file->f_path.dentry, entry);
513 status = nfs_readdir_add_to_array(entry, page);
516 } while (!entry->eof);
518 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
519 array = nfs_readdir_get_array(page);
520 if (!IS_ERR(array)) {
521 array->eof_index = array->size;
523 nfs_readdir_release_array(page);
525 status = PTR_ERR(array);
533 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
536 for (i = 0; i < npages; i++)
541 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
544 nfs_readdir_free_pagearray(pages, npages);
548 * nfs_readdir_large_page will allocate pages that must be freed with a call
549 * to nfs_readdir_free_large_page
552 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
556 for (i = 0; i < npages; i++) {
557 struct page *page = alloc_page(GFP_KERNEL);
565 nfs_readdir_free_pagearray(pages, i);
570 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
572 struct page *pages[NFS_MAX_READDIR_PAGES];
573 void *pages_ptr = NULL;
574 struct nfs_entry entry;
575 struct file *file = desc->file;
576 struct nfs_cache_array *array;
577 int status = -ENOMEM;
578 unsigned int array_size = ARRAY_SIZE(pages);
580 entry.prev_cookie = 0;
581 entry.cookie = desc->last_cookie;
583 entry.fh = nfs_alloc_fhandle();
584 entry.fattr = nfs_alloc_fattr();
585 entry.server = NFS_SERVER(inode);
586 if (entry.fh == NULL || entry.fattr == NULL)
589 array = nfs_readdir_get_array(page);
591 status = PTR_ERR(array);
594 memset(array, 0, sizeof(struct nfs_cache_array));
595 array->eof_index = -1;
597 status = nfs_readdir_large_page(pages, array_size);
599 goto out_release_array;
602 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
607 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
609 if (status == -ENOSPC)
613 } while (array->eof_index < 0);
615 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
617 nfs_readdir_release_array(page);
619 nfs_free_fattr(entry.fattr);
620 nfs_free_fhandle(entry.fh);
625 * Now we cache directories properly, by converting xdr information
626 * to an array that can be used for lookups later. This results in
627 * fewer cache pages, since we can store more information on each page.
628 * We only need to convert from xdr once so future lookups are much simpler
631 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
633 struct inode *inode = file_inode(desc->file);
636 ret = nfs_readdir_xdr_to_array(desc, page, inode);
639 SetPageUptodate(page);
641 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
642 /* Should never happen */
643 nfs_zap_mapping(inode, inode->i_mapping);
653 void cache_page_release(nfs_readdir_descriptor_t *desc)
655 if (!desc->page->mapping)
656 nfs_readdir_clear_array(desc->page);
657 page_cache_release(desc->page);
662 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
664 return read_cache_page(file_inode(desc->file)->i_mapping,
665 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
669 * Returns 0 if desc->dir_cookie was found on page desc->page_index
672 int find_cache_page(nfs_readdir_descriptor_t *desc)
676 desc->page = get_cache_page(desc);
677 if (IS_ERR(desc->page))
678 return PTR_ERR(desc->page);
680 res = nfs_readdir_search_array(desc);
682 cache_page_release(desc);
686 /* Search for desc->dir_cookie from the beginning of the page cache */
688 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
692 if (desc->page_index == 0) {
693 desc->current_index = 0;
694 desc->last_cookie = 0;
697 res = find_cache_page(desc);
698 } while (res == -EAGAIN);
703 * Once we've found the start of the dirent within a page: fill 'er up...
706 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
708 struct file *file = desc->file;
711 struct nfs_cache_array *array = NULL;
712 struct nfs_open_dir_context *ctx = file->private_data;
714 array = nfs_readdir_get_array(desc->page);
716 res = PTR_ERR(array);
720 for (i = desc->cache_entry_index; i < array->size; i++) {
721 struct nfs_cache_array_entry *ent;
723 ent = &array->array[i];
724 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
725 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
730 if (i < (array->size-1))
731 *desc->dir_cookie = array->array[i+1].cookie;
733 *desc->dir_cookie = array->last_cookie;
737 if (array->eof_index >= 0)
740 nfs_readdir_release_array(desc->page);
742 cache_page_release(desc);
743 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
744 (unsigned long long)*desc->dir_cookie, res);
749 * If we cannot find a cookie in our cache, we suspect that this is
750 * because it points to a deleted file, so we ask the server to return
751 * whatever it thinks is the next entry. We then feed this to filldir.
752 * If all goes well, we should then be able to find our way round the
753 * cache on the next call to readdir_search_pagecache();
755 * NOTE: we cannot add the anonymous page to the pagecache because
756 * the data it contains might not be page aligned. Besides,
757 * we should already have a complete representation of the
758 * directory in the page cache by the time we get here.
761 int uncached_readdir(nfs_readdir_descriptor_t *desc)
763 struct page *page = NULL;
765 struct inode *inode = file_inode(desc->file);
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);
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 *file, struct dir_context *ctx)
803 struct dentry *dentry = file->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 = file->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)ctx->pos);
813 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
816 * ctx->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));
825 desc->dir_cookie = &dir_ctx->dir_cookie;
826 desc->decode = NFS_PROTO(inode)->decode_dirent;
827 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
829 nfs_block_sillyrename(dentry);
830 res = nfs_revalidate_mapping(inode, file->f_mapping);
835 res = readdir_search_pagecache(desc);
837 if (res == -EBADCOOKIE) {
839 /* This means either end of directory */
840 if (*desc->dir_cookie && desc->eof == 0) {
841 /* Or that the server has 'lost' a cookie */
842 res = uncached_readdir(desc);
848 if (res == -ETOOSMALL && desc->plus) {
849 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
850 nfs_zap_caches(inode);
851 desc->page_index = 0;
859 res = nfs_do_filldir(desc);
862 } while (!desc->eof);
864 nfs_unblock_sillyrename(dentry);
867 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
868 dentry->d_parent->d_name.name, dentry->d_name.name,
873 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
875 struct dentry *dentry = filp->f_path.dentry;
876 struct inode *inode = dentry->d_inode;
877 struct nfs_open_dir_context *dir_ctx = filp->private_data;
879 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
880 dentry->d_parent->d_name.name,
884 mutex_lock(&inode->i_mutex);
887 offset += filp->f_pos;
895 if (offset != filp->f_pos) {
896 filp->f_pos = offset;
897 dir_ctx->dir_cookie = 0;
901 mutex_unlock(&inode->i_mutex);
906 * All directory operations under NFS are synchronous, so fsync()
907 * is a dummy operation.
909 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
912 struct dentry *dentry = filp->f_path.dentry;
913 struct inode *inode = dentry->d_inode;
915 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
916 dentry->d_parent->d_name.name, dentry->d_name.name,
919 mutex_lock(&inode->i_mutex);
920 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
921 mutex_unlock(&inode->i_mutex);
926 * nfs_force_lookup_revalidate - Mark the directory as having changed
927 * @dir - pointer to directory inode
929 * This forces the revalidation code in nfs_lookup_revalidate() to do a
930 * full lookup on all child dentries of 'dir' whenever a change occurs
931 * on the server that might have invalidated our dcache.
933 * The caller should be holding dir->i_lock
935 void nfs_force_lookup_revalidate(struct inode *dir)
937 NFS_I(dir)->cache_change_attribute++;
939 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
942 * A check for whether or not the parent directory has changed.
943 * In the case it has, we assume that the dentries are untrustworthy
944 * and may need to be looked up again.
946 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
950 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
952 if (!nfs_verify_change_attribute(dir, dentry->d_time))
954 /* Revalidate nfsi->cache_change_attribute before we declare a match */
955 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
957 if (!nfs_verify_change_attribute(dir, dentry->d_time))
963 * Use intent information to check whether or not we're going to do
964 * an O_EXCL create using this path component.
966 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
968 if (NFS_PROTO(dir)->version == 2)
970 return flags & LOOKUP_EXCL;
974 * Inode and filehandle revalidation for lookups.
976 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
977 * or if the intent information indicates that we're about to open this
978 * particular file and the "nocto" mount flag is not set.
982 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
984 struct nfs_server *server = NFS_SERVER(inode);
987 if (IS_AUTOMOUNT(inode))
989 /* VFS wants an on-the-wire revalidation */
990 if (flags & LOOKUP_REVAL)
992 /* This is an open(2) */
993 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
994 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
997 return (inode->i_nlink == 0) ? -ENOENT : 0;
999 ret = __nfs_revalidate_inode(server, inode);
1006 * We judge how long we want to trust negative
1007 * dentries by looking at the parent inode mtime.
1009 * If parent mtime has changed, we revalidate, else we wait for a
1010 * period corresponding to the parent's attribute cache timeout value.
1013 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1016 /* Don't revalidate a negative dentry if we're creating a new file */
1017 if (flags & LOOKUP_CREATE)
1019 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1021 return !nfs_check_verifier(dir, dentry);
1025 * This is called every time the dcache has a lookup hit,
1026 * and we should check whether we can really trust that
1029 * NOTE! The hit can be a negative hit too, don't assume
1032 * If the parent directory is seen to have changed, we throw out the
1033 * cached dentry and do a new lookup.
1035 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1038 struct inode *inode;
1039 struct dentry *parent;
1040 struct nfs_fh *fhandle = NULL;
1041 struct nfs_fattr *fattr = NULL;
1044 if (flags & LOOKUP_RCU)
1047 parent = dget_parent(dentry);
1048 dir = parent->d_inode;
1049 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1050 inode = dentry->d_inode;
1053 if (nfs_neg_need_reval(dir, dentry, flags))
1055 goto out_valid_noent;
1058 if (is_bad_inode(inode)) {
1059 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1060 __func__, dentry->d_parent->d_name.name,
1061 dentry->d_name.name);
1065 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1066 goto out_set_verifier;
1068 /* Force a full look up iff the parent directory has changed */
1069 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1070 if (nfs_lookup_verify_inode(inode, flags))
1071 goto out_zap_parent;
1075 if (NFS_STALE(inode))
1079 fhandle = nfs_alloc_fhandle();
1080 fattr = nfs_alloc_fattr();
1081 if (fhandle == NULL || fattr == NULL)
1084 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1087 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1089 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1092 nfs_free_fattr(fattr);
1093 nfs_free_fhandle(fhandle);
1095 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1097 /* Success: notify readdir to use READDIRPLUS */
1098 nfs_advise_use_readdirplus(dir);
1101 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1102 __func__, dentry->d_parent->d_name.name,
1103 dentry->d_name.name);
1106 nfs_zap_caches(dir);
1108 nfs_free_fattr(fattr);
1109 nfs_free_fhandle(fhandle);
1110 nfs_mark_for_revalidate(dir);
1111 if (inode && S_ISDIR(inode->i_mode)) {
1112 /* Purge readdir caches. */
1113 nfs_zap_caches(inode);
1114 /* If we have submounts, don't unhash ! */
1115 if (have_submounts(dentry))
1117 if (dentry->d_flags & DCACHE_DISCONNECTED)
1119 shrink_dcache_parent(dentry);
1123 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1124 __func__, dentry->d_parent->d_name.name,
1125 dentry->d_name.name);
1128 nfs_free_fattr(fattr);
1129 nfs_free_fhandle(fhandle);
1131 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1132 __func__, dentry->d_parent->d_name.name,
1133 dentry->d_name.name, error);
1138 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1139 * when we don't really care about the dentry name. This is called when a
1140 * pathwalk ends on a dentry that was not found via a normal lookup in the
1141 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1143 * In this situation, we just want to verify that the inode itself is OK
1144 * since the dentry might have changed on the server.
1146 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1149 struct inode *inode = dentry->d_inode;
1152 * I believe we can only get a negative dentry here in the case of a
1153 * procfs-style symlink. Just assume it's correct for now, but we may
1154 * eventually need to do something more here.
1157 dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1158 __func__, dentry->d_parent->d_name.name,
1159 dentry->d_name.name);
1163 if (is_bad_inode(inode)) {
1164 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1165 __func__, dentry->d_parent->d_name.name,
1166 dentry->d_name.name);
1170 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1171 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1172 __func__, inode->i_ino, error ? "invalid" : "valid");
1177 * This is called from dput() when d_count is going to 0.
1179 static int nfs_dentry_delete(const struct dentry *dentry)
1181 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1182 dentry->d_parent->d_name.name, dentry->d_name.name,
1185 /* Unhash any dentry with a stale inode */
1186 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1189 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1190 /* Unhash it, so that ->d_iput() would be called */
1193 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1194 /* Unhash it, so that ancestors of killed async unlink
1195 * files will be cleaned up during umount */
1202 /* Ensure that we revalidate inode->i_nlink */
1203 static void nfs_drop_nlink(struct inode *inode)
1205 spin_lock(&inode->i_lock);
1206 /* drop the inode if we're reasonably sure this is the last link */
1207 if (inode->i_nlink == 1)
1209 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1210 spin_unlock(&inode->i_lock);
1214 * Called when the dentry loses inode.
1215 * We use it to clean up silly-renamed files.
1217 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1219 if (S_ISDIR(inode->i_mode))
1220 /* drop any readdir cache as it could easily be old */
1221 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1223 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1224 nfs_complete_unlink(dentry, inode);
1225 nfs_drop_nlink(inode);
1230 static void nfs_d_release(struct dentry *dentry)
1232 /* free cached devname value, if it survived that far */
1233 if (unlikely(dentry->d_fsdata)) {
1234 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1237 kfree(dentry->d_fsdata);
1241 const struct dentry_operations nfs_dentry_operations = {
1242 .d_revalidate = nfs_lookup_revalidate,
1243 .d_weak_revalidate = nfs_weak_revalidate,
1244 .d_delete = nfs_dentry_delete,
1245 .d_iput = nfs_dentry_iput,
1246 .d_automount = nfs_d_automount,
1247 .d_release = nfs_d_release,
1249 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1251 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1254 struct dentry *parent;
1255 struct inode *inode = NULL;
1256 struct nfs_fh *fhandle = NULL;
1257 struct nfs_fattr *fattr = NULL;
1260 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1261 dentry->d_parent->d_name.name, dentry->d_name.name);
1262 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1264 res = ERR_PTR(-ENAMETOOLONG);
1265 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1269 * If we're doing an exclusive create, optimize away the lookup
1270 * but don't hash the dentry.
1272 if (nfs_is_exclusive_create(dir, flags)) {
1273 d_instantiate(dentry, NULL);
1278 res = ERR_PTR(-ENOMEM);
1279 fhandle = nfs_alloc_fhandle();
1280 fattr = nfs_alloc_fattr();
1281 if (fhandle == NULL || fattr == NULL)
1284 parent = dentry->d_parent;
1285 /* Protect against concurrent sillydeletes */
1286 nfs_block_sillyrename(parent);
1287 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1288 if (error == -ENOENT)
1291 res = ERR_PTR(error);
1292 goto out_unblock_sillyrename;
1294 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1295 res = ERR_CAST(inode);
1297 goto out_unblock_sillyrename;
1299 /* Success: notify readdir to use READDIRPLUS */
1300 nfs_advise_use_readdirplus(dir);
1303 res = d_materialise_unique(dentry, inode);
1306 goto out_unblock_sillyrename;
1309 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1310 out_unblock_sillyrename:
1311 nfs_unblock_sillyrename(parent);
1313 nfs_free_fattr(fattr);
1314 nfs_free_fhandle(fhandle);
1317 EXPORT_SYMBOL_GPL(nfs_lookup);
1319 #if IS_ENABLED(CONFIG_NFS_V4)
1320 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1322 const struct dentry_operations nfs4_dentry_operations = {
1323 .d_revalidate = nfs4_lookup_revalidate,
1324 .d_delete = nfs_dentry_delete,
1325 .d_iput = nfs_dentry_iput,
1326 .d_automount = nfs_d_automount,
1327 .d_release = nfs_d_release,
1329 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1331 static fmode_t flags_to_mode(int flags)
1333 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1334 if ((flags & O_ACCMODE) != O_WRONLY)
1336 if ((flags & O_ACCMODE) != O_RDONLY)
1341 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1343 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1346 static int do_open(struct inode *inode, struct file *filp)
1348 nfs_fscache_set_inode_cookie(inode, filp);
1352 static int nfs_finish_open(struct nfs_open_context *ctx,
1353 struct dentry *dentry,
1354 struct file *file, unsigned open_flags,
1359 if (ctx->dentry != dentry) {
1361 ctx->dentry = dget(dentry);
1364 /* If the open_intent is for execute, we have an extra check to make */
1365 if (ctx->mode & FMODE_EXEC) {
1366 err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1371 err = finish_open(file, dentry, do_open, opened);
1374 nfs_file_set_open_context(file, ctx);
1377 put_nfs_open_context(ctx);
1381 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1382 struct file *file, unsigned open_flags,
1383 umode_t mode, int *opened)
1385 struct nfs_open_context *ctx;
1387 struct iattr attr = { .ia_valid = ATTR_OPEN };
1388 struct inode *inode;
1391 /* Expect a negative dentry */
1392 BUG_ON(dentry->d_inode);
1394 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1395 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1397 /* NFS only supports OPEN on regular files */
1398 if ((open_flags & O_DIRECTORY)) {
1399 if (!d_unhashed(dentry)) {
1401 * Hashed negative dentry with O_DIRECTORY: dentry was
1402 * revalidated and is fine, no need to perform lookup
1410 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1411 return -ENAMETOOLONG;
1413 if (open_flags & O_CREAT) {
1414 attr.ia_valid |= ATTR_MODE;
1415 attr.ia_mode = mode & ~current_umask();
1417 if (open_flags & O_TRUNC) {
1418 attr.ia_valid |= ATTR_SIZE;
1422 ctx = create_nfs_open_context(dentry, open_flags);
1427 nfs_block_sillyrename(dentry->d_parent);
1428 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1430 if (IS_ERR(inode)) {
1431 nfs_unblock_sillyrename(dentry->d_parent);
1432 put_nfs_open_context(ctx);
1433 err = PTR_ERR(inode);
1436 d_add(dentry, NULL);
1442 if (!(open_flags & O_NOFOLLOW))
1451 res = d_add_unique(dentry, inode);
1455 nfs_unblock_sillyrename(dentry->d_parent);
1456 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1458 err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1465 res = nfs_lookup(dir, dentry, 0);
1470 return finish_no_open(file, res);
1472 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1474 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1476 struct dentry *parent = NULL;
1477 struct inode *inode;
1481 if (flags & LOOKUP_RCU)
1484 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1486 if (d_mountpoint(dentry))
1488 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1491 inode = dentry->d_inode;
1492 parent = dget_parent(dentry);
1493 dir = parent->d_inode;
1495 /* We can't create new files in nfs_open_revalidate(), so we
1496 * optimize away revalidation of negative dentries.
1498 if (inode == NULL) {
1499 if (!nfs_neg_need_reval(dir, dentry, flags))
1504 /* NFS only supports OPEN on regular files */
1505 if (!S_ISREG(inode->i_mode))
1507 /* We cannot do exclusive creation on a positive dentry */
1508 if (flags & LOOKUP_EXCL)
1511 /* Let f_op->open() actually open (and revalidate) the file */
1521 return nfs_lookup_revalidate(dentry, flags);
1524 #endif /* CONFIG_NFSV4 */
1527 * Code common to create, mkdir, and mknod.
1529 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1530 struct nfs_fattr *fattr)
1532 struct dentry *parent = dget_parent(dentry);
1533 struct inode *dir = parent->d_inode;
1534 struct inode *inode;
1535 int error = -EACCES;
1539 /* We may have been initialized further down */
1540 if (dentry->d_inode)
1542 if (fhandle->size == 0) {
1543 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1547 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1548 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1549 struct nfs_server *server = NFS_SB(dentry->d_sb);
1550 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1554 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1555 error = PTR_ERR(inode);
1558 d_add(dentry, inode);
1563 nfs_mark_for_revalidate(dir);
1567 EXPORT_SYMBOL_GPL(nfs_instantiate);
1570 * Following a failed create operation, we drop the dentry rather
1571 * than retain a negative dentry. This avoids a problem in the event
1572 * that the operation succeeded on the server, but an error in the
1573 * reply path made it appear to have failed.
1575 int nfs_create(struct inode *dir, struct dentry *dentry,
1576 umode_t mode, bool excl)
1579 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1582 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1583 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1585 attr.ia_mode = mode;
1586 attr.ia_valid = ATTR_MODE;
1588 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1596 EXPORT_SYMBOL_GPL(nfs_create);
1599 * See comments for nfs_proc_create regarding failed operations.
1602 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1607 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1608 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1610 if (!new_valid_dev(rdev))
1613 attr.ia_mode = mode;
1614 attr.ia_valid = ATTR_MODE;
1616 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1624 EXPORT_SYMBOL_GPL(nfs_mknod);
1627 * See comments for nfs_proc_create regarding failed operations.
1629 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1634 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1635 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1637 attr.ia_valid = ATTR_MODE;
1638 attr.ia_mode = mode | S_IFDIR;
1640 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1648 EXPORT_SYMBOL_GPL(nfs_mkdir);
1650 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1652 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1656 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1660 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1661 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1663 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1664 /* Ensure the VFS deletes this inode */
1665 if (error == 0 && dentry->d_inode != NULL)
1666 clear_nlink(dentry->d_inode);
1667 else if (error == -ENOENT)
1668 nfs_dentry_handle_enoent(dentry);
1672 EXPORT_SYMBOL_GPL(nfs_rmdir);
1675 * Remove a file after making sure there are no pending writes,
1676 * and after checking that the file has only one user.
1678 * We invalidate the attribute cache and free the inode prior to the operation
1679 * to avoid possible races if the server reuses the inode.
1681 static int nfs_safe_remove(struct dentry *dentry)
1683 struct inode *dir = dentry->d_parent->d_inode;
1684 struct inode *inode = dentry->d_inode;
1687 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1688 dentry->d_parent->d_name.name, dentry->d_name.name);
1690 /* If the dentry was sillyrenamed, we simply call d_delete() */
1691 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1696 if (inode != NULL) {
1697 NFS_PROTO(inode)->return_delegation(inode);
1698 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1700 nfs_drop_nlink(inode);
1702 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1703 if (error == -ENOENT)
1704 nfs_dentry_handle_enoent(dentry);
1709 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1710 * belongs to an active ".nfs..." file and we return -EBUSY.
1712 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1714 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1717 int need_rehash = 0;
1719 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1720 dir->i_ino, dentry->d_name.name);
1722 spin_lock(&dentry->d_lock);
1723 if (dentry->d_count > 1) {
1724 spin_unlock(&dentry->d_lock);
1725 /* Start asynchronous writeout of the inode */
1726 write_inode_now(dentry->d_inode, 0);
1727 error = nfs_sillyrename(dir, dentry);
1730 if (!d_unhashed(dentry)) {
1734 spin_unlock(&dentry->d_lock);
1735 error = nfs_safe_remove(dentry);
1736 if (!error || error == -ENOENT) {
1737 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1738 } else if (need_rehash)
1742 EXPORT_SYMBOL_GPL(nfs_unlink);
1745 * To create a symbolic link, most file systems instantiate a new inode,
1746 * add a page to it containing the path, then write it out to the disk
1747 * using prepare_write/commit_write.
1749 * Unfortunately the NFS client can't create the in-core inode first
1750 * because it needs a file handle to create an in-core inode (see
1751 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1752 * symlink request has completed on the server.
1754 * So instead we allocate a raw page, copy the symname into it, then do
1755 * the SYMLINK request with the page as the buffer. If it succeeds, we
1756 * now have a new file handle and can instantiate an in-core NFS inode
1757 * and move the raw page into its mapping.
1759 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1761 struct pagevec lru_pvec;
1765 unsigned int pathlen = strlen(symname);
1768 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1769 dir->i_ino, dentry->d_name.name, symname);
1771 if (pathlen > PAGE_SIZE)
1772 return -ENAMETOOLONG;
1774 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1775 attr.ia_valid = ATTR_MODE;
1777 page = alloc_page(GFP_HIGHUSER);
1781 kaddr = kmap_atomic(page);
1782 memcpy(kaddr, symname, pathlen);
1783 if (pathlen < PAGE_SIZE)
1784 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1785 kunmap_atomic(kaddr);
1787 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1789 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1790 dir->i_sb->s_id, dir->i_ino,
1791 dentry->d_name.name, symname, error);
1798 * No big deal if we can't add this page to the page cache here.
1799 * READLINK will get the missing page from the server if needed.
1801 pagevec_init(&lru_pvec, 0);
1802 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1804 pagevec_add(&lru_pvec, page);
1805 pagevec_lru_add_file(&lru_pvec);
1806 SetPageUptodate(page);
1813 EXPORT_SYMBOL_GPL(nfs_symlink);
1816 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1818 struct inode *inode = old_dentry->d_inode;
1821 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1822 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1823 dentry->d_parent->d_name.name, dentry->d_name.name);
1825 NFS_PROTO(inode)->return_delegation(inode);
1828 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1831 d_add(dentry, inode);
1835 EXPORT_SYMBOL_GPL(nfs_link);
1839 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1840 * different file handle for the same inode after a rename (e.g. when
1841 * moving to a different directory). A fail-safe method to do so would
1842 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1843 * rename the old file using the sillyrename stuff. This way, the original
1844 * file in old_dir will go away when the last process iput()s the inode.
1848 * It actually works quite well. One needs to have the possibility for
1849 * at least one ".nfs..." file in each directory the file ever gets
1850 * moved or linked to which happens automagically with the new
1851 * implementation that only depends on the dcache stuff instead of
1852 * using the inode layer
1854 * Unfortunately, things are a little more complicated than indicated
1855 * above. For a cross-directory move, we want to make sure we can get
1856 * rid of the old inode after the operation. This means there must be
1857 * no pending writes (if it's a file), and the use count must be 1.
1858 * If these conditions are met, we can drop the dentries before doing
1861 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1862 struct inode *new_dir, struct dentry *new_dentry)
1864 struct inode *old_inode = old_dentry->d_inode;
1865 struct inode *new_inode = new_dentry->d_inode;
1866 struct dentry *dentry = NULL, *rehash = NULL;
1869 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1870 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1871 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1872 new_dentry->d_count);
1875 * For non-directories, check whether the target is busy and if so,
1876 * make a copy of the dentry and then do a silly-rename. If the
1877 * silly-rename succeeds, the copied dentry is hashed and becomes
1880 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1882 * To prevent any new references to the target during the
1883 * rename, we unhash the dentry in advance.
1885 if (!d_unhashed(new_dentry)) {
1887 rehash = new_dentry;
1890 if (new_dentry->d_count > 2) {
1893 /* copy the target dentry's name */
1894 dentry = d_alloc(new_dentry->d_parent,
1895 &new_dentry->d_name);
1899 /* silly-rename the existing target ... */
1900 err = nfs_sillyrename(new_dir, new_dentry);
1904 new_dentry = dentry;
1910 NFS_PROTO(old_inode)->return_delegation(old_inode);
1911 if (new_inode != NULL)
1912 NFS_PROTO(new_inode)->return_delegation(new_inode);
1914 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1915 new_dir, &new_dentry->d_name);
1916 nfs_mark_for_revalidate(old_inode);
1921 if (new_inode != NULL)
1922 nfs_drop_nlink(new_inode);
1923 d_move(old_dentry, new_dentry);
1924 nfs_set_verifier(new_dentry,
1925 nfs_save_change_attribute(new_dir));
1926 } else if (error == -ENOENT)
1927 nfs_dentry_handle_enoent(old_dentry);
1929 /* new dentry created? */
1934 EXPORT_SYMBOL_GPL(nfs_rename);
1936 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1937 static LIST_HEAD(nfs_access_lru_list);
1938 static atomic_long_t nfs_access_nr_entries;
1940 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1942 put_rpccred(entry->cred);
1944 smp_mb__before_atomic_dec();
1945 atomic_long_dec(&nfs_access_nr_entries);
1946 smp_mb__after_atomic_dec();
1949 static void nfs_access_free_list(struct list_head *head)
1951 struct nfs_access_entry *cache;
1953 while (!list_empty(head)) {
1954 cache = list_entry(head->next, struct nfs_access_entry, lru);
1955 list_del(&cache->lru);
1956 nfs_access_free_entry(cache);
1960 int nfs_access_cache_shrinker(struct shrinker *shrink,
1961 struct shrink_control *sc)
1964 struct nfs_inode *nfsi, *next;
1965 struct nfs_access_entry *cache;
1966 int nr_to_scan = sc->nr_to_scan;
1967 gfp_t gfp_mask = sc->gfp_mask;
1969 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1970 return (nr_to_scan == 0) ? 0 : -1;
1972 spin_lock(&nfs_access_lru_lock);
1973 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1974 struct inode *inode;
1976 if (nr_to_scan-- == 0)
1978 inode = &nfsi->vfs_inode;
1979 spin_lock(&inode->i_lock);
1980 if (list_empty(&nfsi->access_cache_entry_lru))
1981 goto remove_lru_entry;
1982 cache = list_entry(nfsi->access_cache_entry_lru.next,
1983 struct nfs_access_entry, lru);
1984 list_move(&cache->lru, &head);
1985 rb_erase(&cache->rb_node, &nfsi->access_cache);
1986 if (!list_empty(&nfsi->access_cache_entry_lru))
1987 list_move_tail(&nfsi->access_cache_inode_lru,
1988 &nfs_access_lru_list);
1991 list_del_init(&nfsi->access_cache_inode_lru);
1992 smp_mb__before_clear_bit();
1993 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1994 smp_mb__after_clear_bit();
1996 spin_unlock(&inode->i_lock);
1998 spin_unlock(&nfs_access_lru_lock);
1999 nfs_access_free_list(&head);
2000 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2003 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2005 struct rb_root *root_node = &nfsi->access_cache;
2007 struct nfs_access_entry *entry;
2009 /* Unhook entries from the cache */
2010 while ((n = rb_first(root_node)) != NULL) {
2011 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2012 rb_erase(n, root_node);
2013 list_move(&entry->lru, head);
2015 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2018 void nfs_access_zap_cache(struct inode *inode)
2022 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2024 /* Remove from global LRU init */
2025 spin_lock(&nfs_access_lru_lock);
2026 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2027 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2029 spin_lock(&inode->i_lock);
2030 __nfs_access_zap_cache(NFS_I(inode), &head);
2031 spin_unlock(&inode->i_lock);
2032 spin_unlock(&nfs_access_lru_lock);
2033 nfs_access_free_list(&head);
2035 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2037 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2039 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2040 struct nfs_access_entry *entry;
2043 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2045 if (cred < entry->cred)
2047 else if (cred > entry->cred)
2055 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2057 struct nfs_inode *nfsi = NFS_I(inode);
2058 struct nfs_access_entry *cache;
2061 spin_lock(&inode->i_lock);
2062 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2064 cache = nfs_access_search_rbtree(inode, cred);
2067 if (!nfs_have_delegated_attributes(inode) &&
2068 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2070 res->jiffies = cache->jiffies;
2071 res->cred = cache->cred;
2072 res->mask = cache->mask;
2073 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2076 spin_unlock(&inode->i_lock);
2079 rb_erase(&cache->rb_node, &nfsi->access_cache);
2080 list_del(&cache->lru);
2081 spin_unlock(&inode->i_lock);
2082 nfs_access_free_entry(cache);
2085 spin_unlock(&inode->i_lock);
2086 nfs_access_zap_cache(inode);
2090 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2092 struct nfs_inode *nfsi = NFS_I(inode);
2093 struct rb_root *root_node = &nfsi->access_cache;
2094 struct rb_node **p = &root_node->rb_node;
2095 struct rb_node *parent = NULL;
2096 struct nfs_access_entry *entry;
2098 spin_lock(&inode->i_lock);
2099 while (*p != NULL) {
2101 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2103 if (set->cred < entry->cred)
2104 p = &parent->rb_left;
2105 else if (set->cred > entry->cred)
2106 p = &parent->rb_right;
2110 rb_link_node(&set->rb_node, parent, p);
2111 rb_insert_color(&set->rb_node, root_node);
2112 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2113 spin_unlock(&inode->i_lock);
2116 rb_replace_node(parent, &set->rb_node, root_node);
2117 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2118 list_del(&entry->lru);
2119 spin_unlock(&inode->i_lock);
2120 nfs_access_free_entry(entry);
2123 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2125 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2128 RB_CLEAR_NODE(&cache->rb_node);
2129 cache->jiffies = set->jiffies;
2130 cache->cred = get_rpccred(set->cred);
2131 cache->mask = set->mask;
2133 nfs_access_add_rbtree(inode, cache);
2135 /* Update accounting */
2136 smp_mb__before_atomic_inc();
2137 atomic_long_inc(&nfs_access_nr_entries);
2138 smp_mb__after_atomic_inc();
2140 /* Add inode to global LRU list */
2141 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2142 spin_lock(&nfs_access_lru_lock);
2143 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2144 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2145 &nfs_access_lru_list);
2146 spin_unlock(&nfs_access_lru_lock);
2149 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2151 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2154 if (access_result & NFS4_ACCESS_READ)
2155 entry->mask |= MAY_READ;
2157 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2158 entry->mask |= MAY_WRITE;
2159 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2160 entry->mask |= MAY_EXEC;
2162 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2164 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2166 struct nfs_access_entry cache;
2169 status = nfs_access_get_cached(inode, cred, &cache);
2173 /* Be clever: ask server to check for all possible rights */
2174 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2176 cache.jiffies = jiffies;
2177 status = NFS_PROTO(inode)->access(inode, &cache);
2179 if (status == -ESTALE) {
2180 nfs_zap_caches(inode);
2181 if (!S_ISDIR(inode->i_mode))
2182 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2186 nfs_access_add_cache(inode, &cache);
2188 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2193 static int nfs_open_permission_mask(int openflags)
2197 if (openflags & __FMODE_EXEC) {
2198 /* ONLY check exec rights */
2201 if ((openflags & O_ACCMODE) != O_WRONLY)
2203 if ((openflags & O_ACCMODE) != O_RDONLY)
2210 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2212 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2214 EXPORT_SYMBOL_GPL(nfs_may_open);
2216 int nfs_permission(struct inode *inode, int mask)
2218 struct rpc_cred *cred;
2221 if (mask & MAY_NOT_BLOCK)
2224 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2226 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2228 /* Is this sys_access() ? */
2229 if (mask & (MAY_ACCESS | MAY_CHDIR))
2232 switch (inode->i_mode & S_IFMT) {
2236 /* NFSv4 has atomic_open... */
2237 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2238 && (mask & MAY_OPEN)
2239 && !(mask & MAY_EXEC))
2244 * Optimize away all write operations, since the server
2245 * will check permissions when we perform the op.
2247 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2252 if (!NFS_PROTO(inode)->access)
2255 cred = rpc_lookup_cred();
2256 if (!IS_ERR(cred)) {
2257 res = nfs_do_access(inode, cred, mask);
2260 res = PTR_ERR(cred);
2262 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2265 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2266 inode->i_sb->s_id, inode->i_ino, mask, res);
2269 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2271 res = generic_permission(inode, mask);
2274 EXPORT_SYMBOL_GPL(nfs_permission);
2278 * version-control: t
2279 * kept-new-versions: 5