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/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
41 #include "delegation.h"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
57 const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .iterate = nfs_readdir,
62 .release = nfs_closedir,
63 .fsync = nfs_fsync_dir,
66 const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
72 struct nfs_open_dir_context *ctx;
73 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
76 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
79 ctx->cred = get_rpccred(cred);
82 return ERR_PTR(-ENOMEM);
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
87 put_rpccred(ctx->cred);
95 nfs_opendir(struct inode *inode, struct file *filp)
98 struct nfs_open_dir_context *ctx;
99 struct rpc_cred *cred;
101 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
103 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
105 cred = rpc_lookup_cred();
107 return PTR_ERR(cred);
108 ctx = alloc_nfs_open_dir_context(inode, cred);
113 filp->private_data = ctx;
114 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
115 /* This is a mountpoint, so d_revalidate will never
116 * have been called, so we need to refresh the
117 * inode (for close-open consistency) ourselves.
119 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
127 nfs_closedir(struct inode *inode, struct file *filp)
129 put_nfs_open_dir_context(filp->private_data);
133 struct nfs_cache_array_entry {
137 unsigned char d_type;
140 struct nfs_cache_array {
144 struct nfs_cache_array_entry array[0];
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
151 struct dir_context *ctx;
152 unsigned long page_index;
155 loff_t current_index;
156 decode_dirent_t decode;
158 unsigned long timestamp;
159 unsigned long gencount;
160 unsigned int cache_entry_index;
163 } nfs_readdir_descriptor_t;
166 * The caller is responsible for calling nfs_readdir_release_array(page)
169 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
173 return ERR_PTR(-EIO);
176 return ERR_PTR(-ENOMEM);
181 void nfs_readdir_release_array(struct page *page)
187 * we are freeing strings created by nfs_add_to_readdir_array()
190 void nfs_readdir_clear_array(struct page *page)
192 struct nfs_cache_array *array;
195 array = kmap_atomic(page);
196 for (i = 0; i < array->size; i++)
197 kfree(array->array[i].string.name);
198 kunmap_atomic(array);
202 * the caller is responsible for freeing qstr.name
203 * when called by nfs_readdir_add_to_array, the strings will be freed in
204 * nfs_clear_readdir_array()
207 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
210 string->name = kmemdup(name, len, GFP_KERNEL);
211 if (string->name == NULL)
214 * Avoid a kmemleak false positive. The pointer to the name is stored
215 * in a page cache page which kmemleak does not scan.
217 kmemleak_not_leak(string->name);
218 string->hash = full_name_hash(name, len);
223 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
225 struct nfs_cache_array *array = nfs_readdir_get_array(page);
226 struct nfs_cache_array_entry *cache_entry;
230 return PTR_ERR(array);
232 cache_entry = &array->array[array->size];
234 /* Check that this entry lies within the page bounds */
236 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
239 cache_entry->cookie = entry->prev_cookie;
240 cache_entry->ino = entry->ino;
241 cache_entry->d_type = entry->d_type;
242 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
245 array->last_cookie = entry->cookie;
248 array->eof_index = array->size;
250 nfs_readdir_release_array(page);
255 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
257 loff_t diff = desc->ctx->pos - desc->current_index;
262 if (diff >= array->size) {
263 if (array->eof_index >= 0)
268 index = (unsigned int)diff;
269 *desc->dir_cookie = array->array[index].cookie;
270 desc->cache_entry_index = index;
278 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
282 int status = -EAGAIN;
284 for (i = 0; i < array->size; i++) {
285 if (array->array[i].cookie == *desc->dir_cookie) {
286 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
287 struct nfs_open_dir_context *ctx = desc->file->private_data;
289 new_pos = desc->current_index + i;
290 if (ctx->attr_gencount != nfsi->attr_gencount
291 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
293 ctx->attr_gencount = nfsi->attr_gencount;
294 } else if (new_pos < desc->ctx->pos) {
296 && ctx->dup_cookie == *desc->dir_cookie) {
297 if (printk_ratelimit()) {
298 pr_notice("NFS: directory %pD2 contains a readdir loop."
299 "Please contact your server vendor. "
300 "The file: %s has duplicate cookie %llu\n",
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;
441 if (filename.name[0] == '.') {
442 if (filename.len == 1)
444 if (filename.len == 2 && filename.name[1] == '.')
447 filename.hash = full_name_hash(filename.name, filename.len);
449 dentry = d_lookup(parent, &filename);
450 if (dentry != NULL) {
451 if (nfs_same_file(dentry, entry)) {
452 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
453 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
455 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
458 if (d_invalidate(dentry) != 0)
464 dentry = d_alloc(parent, &filename);
468 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
472 alias = d_materialise_unique(dentry, inode);
476 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
479 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
485 /* Perform conversion from xdr to cache array */
487 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
488 struct page **xdr_pages, struct page *page, unsigned int buflen)
490 struct xdr_stream stream;
492 struct page *scratch;
493 struct nfs_cache_array *array;
494 unsigned int count = 0;
497 scratch = alloc_page(GFP_KERNEL);
501 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
502 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
505 status = xdr_decode(desc, entry, &stream);
507 if (status == -EAGAIN)
515 nfs_prime_dcache(desc->file->f_path.dentry, entry);
517 status = nfs_readdir_add_to_array(entry, page);
520 } while (!entry->eof);
522 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
523 array = nfs_readdir_get_array(page);
524 if (!IS_ERR(array)) {
525 array->eof_index = array->size;
527 nfs_readdir_release_array(page);
529 status = PTR_ERR(array);
537 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
540 for (i = 0; i < npages; i++)
545 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
548 nfs_readdir_free_pagearray(pages, npages);
552 * nfs_readdir_large_page will allocate pages that must be freed with a call
553 * to nfs_readdir_free_large_page
556 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
560 for (i = 0; i < npages; i++) {
561 struct page *page = alloc_page(GFP_KERNEL);
569 nfs_readdir_free_pagearray(pages, i);
574 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
576 struct page *pages[NFS_MAX_READDIR_PAGES];
577 void *pages_ptr = NULL;
578 struct nfs_entry entry;
579 struct file *file = desc->file;
580 struct nfs_cache_array *array;
581 int status = -ENOMEM;
582 unsigned int array_size = ARRAY_SIZE(pages);
584 entry.prev_cookie = 0;
585 entry.cookie = desc->last_cookie;
587 entry.fh = nfs_alloc_fhandle();
588 entry.fattr = nfs_alloc_fattr();
589 entry.server = NFS_SERVER(inode);
590 if (entry.fh == NULL || entry.fattr == NULL)
593 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
594 if (IS_ERR(entry.label)) {
595 status = PTR_ERR(entry.label);
599 array = nfs_readdir_get_array(page);
601 status = PTR_ERR(array);
604 memset(array, 0, sizeof(struct nfs_cache_array));
605 array->eof_index = -1;
607 status = nfs_readdir_large_page(pages, array_size);
609 goto out_release_array;
612 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
617 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
619 if (status == -ENOSPC)
623 } while (array->eof_index < 0);
625 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
627 nfs_readdir_release_array(page);
629 nfs4_label_free(entry.label);
631 nfs_free_fattr(entry.fattr);
632 nfs_free_fhandle(entry.fh);
637 * Now we cache directories properly, by converting xdr information
638 * to an array that can be used for lookups later. This results in
639 * fewer cache pages, since we can store more information on each page.
640 * We only need to convert from xdr once so future lookups are much simpler
643 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
645 struct inode *inode = file_inode(desc->file);
648 ret = nfs_readdir_xdr_to_array(desc, page, inode);
651 SetPageUptodate(page);
653 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
654 /* Should never happen */
655 nfs_zap_mapping(inode, inode->i_mapping);
665 void cache_page_release(nfs_readdir_descriptor_t *desc)
667 if (!desc->page->mapping)
668 nfs_readdir_clear_array(desc->page);
669 page_cache_release(desc->page);
674 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
676 return read_cache_page(file_inode(desc->file)->i_mapping,
677 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
681 * Returns 0 if desc->dir_cookie was found on page desc->page_index
684 int find_cache_page(nfs_readdir_descriptor_t *desc)
688 desc->page = get_cache_page(desc);
689 if (IS_ERR(desc->page))
690 return PTR_ERR(desc->page);
692 res = nfs_readdir_search_array(desc);
694 cache_page_release(desc);
698 /* Search for desc->dir_cookie from the beginning of the page cache */
700 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
704 if (desc->page_index == 0) {
705 desc->current_index = 0;
706 desc->last_cookie = 0;
709 res = find_cache_page(desc);
710 } while (res == -EAGAIN);
715 * Once we've found the start of the dirent within a page: fill 'er up...
718 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
720 struct file *file = desc->file;
723 struct nfs_cache_array *array = NULL;
724 struct nfs_open_dir_context *ctx = file->private_data;
726 array = nfs_readdir_get_array(desc->page);
728 res = PTR_ERR(array);
732 for (i = desc->cache_entry_index; i < array->size; i++) {
733 struct nfs_cache_array_entry *ent;
735 ent = &array->array[i];
736 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
737 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
742 if (i < (array->size-1))
743 *desc->dir_cookie = array->array[i+1].cookie;
745 *desc->dir_cookie = array->last_cookie;
749 if (array->eof_index >= 0)
752 nfs_readdir_release_array(desc->page);
754 cache_page_release(desc);
755 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
756 (unsigned long long)*desc->dir_cookie, res);
761 * If we cannot find a cookie in our cache, we suspect that this is
762 * because it points to a deleted file, so we ask the server to return
763 * whatever it thinks is the next entry. We then feed this to filldir.
764 * If all goes well, we should then be able to find our way round the
765 * cache on the next call to readdir_search_pagecache();
767 * NOTE: we cannot add the anonymous page to the pagecache because
768 * the data it contains might not be page aligned. Besides,
769 * we should already have a complete representation of the
770 * directory in the page cache by the time we get here.
773 int uncached_readdir(nfs_readdir_descriptor_t *desc)
775 struct page *page = NULL;
777 struct inode *inode = file_inode(desc->file);
778 struct nfs_open_dir_context *ctx = desc->file->private_data;
780 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
781 (unsigned long long)*desc->dir_cookie);
783 page = alloc_page(GFP_HIGHUSER);
789 desc->page_index = 0;
790 desc->last_cookie = *desc->dir_cookie;
794 status = nfs_readdir_xdr_to_array(desc, page, inode);
798 status = nfs_do_filldir(desc);
801 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
805 cache_page_release(desc);
809 /* The file offset position represents the dirent entry number. A
810 last cookie cache takes care of the common case of reading the
813 static int nfs_readdir(struct file *file, struct dir_context *ctx)
815 struct dentry *dentry = file->f_path.dentry;
816 struct inode *inode = dentry->d_inode;
817 nfs_readdir_descriptor_t my_desc,
819 struct nfs_open_dir_context *dir_ctx = file->private_data;
822 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
823 file, (long long)ctx->pos);
824 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
827 * ctx->pos points to the dirent entry number.
828 * *desc->dir_cookie has the cookie for the next entry. We have
829 * to either find the entry with the appropriate number or
830 * revalidate the cookie.
832 memset(desc, 0, sizeof(*desc));
836 desc->dir_cookie = &dir_ctx->dir_cookie;
837 desc->decode = NFS_PROTO(inode)->decode_dirent;
838 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
840 nfs_block_sillyrename(dentry);
841 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
842 res = nfs_revalidate_mapping(inode, file->f_mapping);
847 res = readdir_search_pagecache(desc);
849 if (res == -EBADCOOKIE) {
851 /* This means either end of directory */
852 if (*desc->dir_cookie && desc->eof == 0) {
853 /* Or that the server has 'lost' a cookie */
854 res = uncached_readdir(desc);
860 if (res == -ETOOSMALL && desc->plus) {
861 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
862 nfs_zap_caches(inode);
863 desc->page_index = 0;
871 res = nfs_do_filldir(desc);
874 } while (!desc->eof);
876 nfs_unblock_sillyrename(dentry);
879 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
883 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
885 struct inode *inode = file_inode(filp);
886 struct nfs_open_dir_context *dir_ctx = filp->private_data;
888 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
889 filp, offset, whence);
891 mutex_lock(&inode->i_mutex);
894 offset += filp->f_pos;
902 if (offset != filp->f_pos) {
903 filp->f_pos = offset;
904 dir_ctx->dir_cookie = 0;
908 mutex_unlock(&inode->i_mutex);
913 * All directory operations under NFS are synchronous, so fsync()
914 * is a dummy operation.
916 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
919 struct inode *inode = file_inode(filp);
921 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
923 mutex_lock(&inode->i_mutex);
924 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
925 mutex_unlock(&inode->i_mutex);
930 * nfs_force_lookup_revalidate - Mark the directory as having changed
931 * @dir - pointer to directory inode
933 * This forces the revalidation code in nfs_lookup_revalidate() to do a
934 * full lookup on all child dentries of 'dir' whenever a change occurs
935 * on the server that might have invalidated our dcache.
937 * The caller should be holding dir->i_lock
939 void nfs_force_lookup_revalidate(struct inode *dir)
941 NFS_I(dir)->cache_change_attribute++;
943 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
946 * A check for whether or not the parent directory has changed.
947 * In the case it has, we assume that the dentries are untrustworthy
948 * and may need to be looked up again.
950 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
954 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
956 if (!nfs_verify_change_attribute(dir, dentry->d_time))
958 /* Revalidate nfsi->cache_change_attribute before we declare a match */
959 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
961 if (!nfs_verify_change_attribute(dir, dentry->d_time))
967 * Use intent information to check whether or not we're going to do
968 * an O_EXCL create using this path component.
970 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
972 if (NFS_PROTO(dir)->version == 2)
974 return flags & LOOKUP_EXCL;
978 * Inode and filehandle revalidation for lookups.
980 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
981 * or if the intent information indicates that we're about to open this
982 * particular file and the "nocto" mount flag is not set.
986 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
988 struct nfs_server *server = NFS_SERVER(inode);
991 if (IS_AUTOMOUNT(inode))
993 /* VFS wants an on-the-wire revalidation */
994 if (flags & LOOKUP_REVAL)
996 /* This is an open(2) */
997 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
998 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1001 return (inode->i_nlink == 0) ? -ENOENT : 0;
1003 ret = __nfs_revalidate_inode(server, inode);
1010 * We judge how long we want to trust negative
1011 * dentries by looking at the parent inode mtime.
1013 * If parent mtime has changed, we revalidate, else we wait for a
1014 * period corresponding to the parent's attribute cache timeout value.
1017 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1020 /* Don't revalidate a negative dentry if we're creating a new file */
1021 if (flags & LOOKUP_CREATE)
1023 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1025 return !nfs_check_verifier(dir, dentry);
1029 * This is called every time the dcache has a lookup hit,
1030 * and we should check whether we can really trust that
1033 * NOTE! The hit can be a negative hit too, don't assume
1036 * If the parent directory is seen to have changed, we throw out the
1037 * cached dentry and do a new lookup.
1039 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1042 struct inode *inode;
1043 struct dentry *parent;
1044 struct nfs_fh *fhandle = NULL;
1045 struct nfs_fattr *fattr = NULL;
1046 struct nfs4_label *label = NULL;
1049 if (flags & LOOKUP_RCU)
1052 parent = dget_parent(dentry);
1053 dir = parent->d_inode;
1054 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1055 inode = dentry->d_inode;
1058 if (nfs_neg_need_reval(dir, dentry, flags))
1060 goto out_valid_noent;
1063 if (is_bad_inode(inode)) {
1064 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1069 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1070 goto out_set_verifier;
1072 /* Force a full look up iff the parent directory has changed */
1073 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1074 if (nfs_lookup_verify_inode(inode, flags))
1075 goto out_zap_parent;
1079 if (NFS_STALE(inode))
1083 fhandle = nfs_alloc_fhandle();
1084 fattr = nfs_alloc_fattr();
1085 if (fhandle == NULL || fattr == NULL)
1088 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1092 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1093 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1094 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1097 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1099 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1102 nfs_setsecurity(inode, fattr, label);
1104 nfs_free_fattr(fattr);
1105 nfs_free_fhandle(fhandle);
1106 nfs4_label_free(label);
1109 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1111 /* Success: notify readdir to use READDIRPLUS */
1112 nfs_advise_use_readdirplus(dir);
1115 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1119 nfs_zap_caches(dir);
1121 nfs_free_fattr(fattr);
1122 nfs_free_fhandle(fhandle);
1123 nfs4_label_free(label);
1124 nfs_mark_for_revalidate(dir);
1125 if (inode && S_ISDIR(inode->i_mode)) {
1126 /* Purge readdir caches. */
1127 nfs_zap_caches(inode);
1128 if (dentry->d_flags & DCACHE_DISCONNECTED)
1131 shrink_submounts_and_drop(dentry);
1133 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1137 nfs_free_fattr(fattr);
1138 nfs_free_fhandle(fhandle);
1139 nfs4_label_free(label);
1141 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1142 __func__, dentry, error);
1147 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1148 * when we don't really care about the dentry name. This is called when a
1149 * pathwalk ends on a dentry that was not found via a normal lookup in the
1150 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1152 * In this situation, we just want to verify that the inode itself is OK
1153 * since the dentry might have changed on the server.
1155 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1158 struct inode *inode = dentry->d_inode;
1161 * I believe we can only get a negative dentry here in the case of a
1162 * procfs-style symlink. Just assume it's correct for now, but we may
1163 * eventually need to do something more here.
1166 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1171 if (is_bad_inode(inode)) {
1172 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1177 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1178 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1179 __func__, inode->i_ino, error ? "invalid" : "valid");
1184 * This is called from dput() when d_count is going to 0.
1186 static int nfs_dentry_delete(const struct dentry *dentry)
1188 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1189 dentry, dentry->d_flags);
1191 /* Unhash any dentry with a stale inode */
1192 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1195 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1196 /* Unhash it, so that ->d_iput() would be called */
1199 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1200 /* Unhash it, so that ancestors of killed async unlink
1201 * files will be cleaned up during umount */
1208 /* Ensure that we revalidate inode->i_nlink */
1209 static void nfs_drop_nlink(struct inode *inode)
1211 spin_lock(&inode->i_lock);
1212 /* drop the inode if we're reasonably sure this is the last link */
1213 if (inode->i_nlink == 1)
1215 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1216 spin_unlock(&inode->i_lock);
1220 * Called when the dentry loses inode.
1221 * We use it to clean up silly-renamed files.
1223 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1225 if (S_ISDIR(inode->i_mode))
1226 /* drop any readdir cache as it could easily be old */
1227 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1229 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1230 nfs_complete_unlink(dentry, inode);
1231 nfs_drop_nlink(inode);
1236 static void nfs_d_release(struct dentry *dentry)
1238 /* free cached devname value, if it survived that far */
1239 if (unlikely(dentry->d_fsdata)) {
1240 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1243 kfree(dentry->d_fsdata);
1247 const struct dentry_operations nfs_dentry_operations = {
1248 .d_revalidate = nfs_lookup_revalidate,
1249 .d_weak_revalidate = nfs_weak_revalidate,
1250 .d_delete = nfs_dentry_delete,
1251 .d_iput = nfs_dentry_iput,
1252 .d_automount = nfs_d_automount,
1253 .d_release = nfs_d_release,
1255 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1257 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1260 struct dentry *parent;
1261 struct inode *inode = NULL;
1262 struct nfs_fh *fhandle = NULL;
1263 struct nfs_fattr *fattr = NULL;
1264 struct nfs4_label *label = NULL;
1267 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1268 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1270 res = ERR_PTR(-ENAMETOOLONG);
1271 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1275 * If we're doing an exclusive create, optimize away the lookup
1276 * but don't hash the dentry.
1278 if (nfs_is_exclusive_create(dir, flags)) {
1279 d_instantiate(dentry, NULL);
1284 res = ERR_PTR(-ENOMEM);
1285 fhandle = nfs_alloc_fhandle();
1286 fattr = nfs_alloc_fattr();
1287 if (fhandle == NULL || fattr == NULL)
1290 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1294 parent = dentry->d_parent;
1295 /* Protect against concurrent sillydeletes */
1296 trace_nfs_lookup_enter(dir, dentry, flags);
1297 nfs_block_sillyrename(parent);
1298 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1299 if (error == -ENOENT)
1302 res = ERR_PTR(error);
1303 goto out_unblock_sillyrename;
1305 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1306 res = ERR_CAST(inode);
1308 goto out_unblock_sillyrename;
1310 /* Success: notify readdir to use READDIRPLUS */
1311 nfs_advise_use_readdirplus(dir);
1314 res = d_materialise_unique(dentry, inode);
1317 goto out_unblock_sillyrename;
1320 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1321 out_unblock_sillyrename:
1322 nfs_unblock_sillyrename(parent);
1323 trace_nfs_lookup_exit(dir, dentry, flags, error);
1324 nfs4_label_free(label);
1326 nfs_free_fattr(fattr);
1327 nfs_free_fhandle(fhandle);
1330 EXPORT_SYMBOL_GPL(nfs_lookup);
1332 #if IS_ENABLED(CONFIG_NFS_V4)
1333 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1335 const struct dentry_operations nfs4_dentry_operations = {
1336 .d_revalidate = nfs4_lookup_revalidate,
1337 .d_delete = nfs_dentry_delete,
1338 .d_iput = nfs_dentry_iput,
1339 .d_automount = nfs_d_automount,
1340 .d_release = nfs_d_release,
1342 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1344 static fmode_t flags_to_mode(int flags)
1346 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1347 if ((flags & O_ACCMODE) != O_WRONLY)
1349 if ((flags & O_ACCMODE) != O_RDONLY)
1354 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1356 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1359 static int do_open(struct inode *inode, struct file *filp)
1361 nfs_fscache_open_file(inode, filp);
1365 static int nfs_finish_open(struct nfs_open_context *ctx,
1366 struct dentry *dentry,
1367 struct file *file, unsigned open_flags,
1372 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1373 *opened |= FILE_CREATED;
1375 err = finish_open(file, dentry, do_open, opened);
1378 nfs_file_set_open_context(file, ctx);
1384 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1385 struct file *file, unsigned open_flags,
1386 umode_t mode, int *opened)
1388 struct nfs_open_context *ctx;
1390 struct iattr attr = { .ia_valid = ATTR_OPEN };
1391 struct inode *inode;
1392 unsigned int lookup_flags = 0;
1395 /* Expect a negative dentry */
1396 BUG_ON(dentry->d_inode);
1398 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %pd\n",
1399 dir->i_sb->s_id, dir->i_ino, dentry);
1401 err = nfs_check_flags(open_flags);
1405 /* NFS only supports OPEN on regular files */
1406 if ((open_flags & O_DIRECTORY)) {
1407 if (!d_unhashed(dentry)) {
1409 * Hashed negative dentry with O_DIRECTORY: dentry was
1410 * revalidated and is fine, no need to perform lookup
1415 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1419 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1420 return -ENAMETOOLONG;
1422 if (open_flags & O_CREAT) {
1423 attr.ia_valid |= ATTR_MODE;
1424 attr.ia_mode = mode & ~current_umask();
1426 if (open_flags & O_TRUNC) {
1427 attr.ia_valid |= ATTR_SIZE;
1431 ctx = create_nfs_open_context(dentry, open_flags);
1436 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1437 nfs_block_sillyrename(dentry->d_parent);
1438 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1439 nfs_unblock_sillyrename(dentry->d_parent);
1440 if (IS_ERR(inode)) {
1441 err = PTR_ERR(inode);
1442 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1443 put_nfs_open_context(ctx);
1447 d_add(dentry, NULL);
1453 if (!(open_flags & O_NOFOLLOW))
1463 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1464 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1465 put_nfs_open_context(ctx);
1470 res = nfs_lookup(dir, dentry, lookup_flags);
1475 return finish_no_open(file, res);
1477 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1479 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1481 struct dentry *parent = NULL;
1482 struct inode *inode;
1486 if (flags & LOOKUP_RCU)
1489 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1491 if (d_mountpoint(dentry))
1493 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1496 inode = dentry->d_inode;
1497 parent = dget_parent(dentry);
1498 dir = parent->d_inode;
1500 /* We can't create new files in nfs_open_revalidate(), so we
1501 * optimize away revalidation of negative dentries.
1503 if (inode == NULL) {
1504 if (!nfs_neg_need_reval(dir, dentry, flags))
1509 /* NFS only supports OPEN on regular files */
1510 if (!S_ISREG(inode->i_mode))
1512 /* We cannot do exclusive creation on a positive dentry */
1513 if (flags & LOOKUP_EXCL)
1516 /* Let f_op->open() actually open (and revalidate) the file */
1526 return nfs_lookup_revalidate(dentry, flags);
1529 #endif /* CONFIG_NFSV4 */
1532 * Code common to create, mkdir, and mknod.
1534 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1535 struct nfs_fattr *fattr,
1536 struct nfs4_label *label)
1538 struct dentry *parent = dget_parent(dentry);
1539 struct inode *dir = parent->d_inode;
1540 struct inode *inode;
1541 int error = -EACCES;
1545 /* We may have been initialized further down */
1546 if (dentry->d_inode)
1548 if (fhandle->size == 0) {
1549 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1553 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1554 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1555 struct nfs_server *server = NFS_SB(dentry->d_sb);
1556 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1560 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1561 error = PTR_ERR(inode);
1564 d_add(dentry, inode);
1569 nfs_mark_for_revalidate(dir);
1573 EXPORT_SYMBOL_GPL(nfs_instantiate);
1576 * Following a failed create operation, we drop the dentry rather
1577 * than retain a negative dentry. This avoids a problem in the event
1578 * that the operation succeeded on the server, but an error in the
1579 * reply path made it appear to have failed.
1581 int nfs_create(struct inode *dir, struct dentry *dentry,
1582 umode_t mode, bool excl)
1585 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1588 dfprintk(VFS, "NFS: create(%s/%ld), %pd\n",
1589 dir->i_sb->s_id, dir->i_ino, dentry);
1591 attr.ia_mode = mode;
1592 attr.ia_valid = ATTR_MODE;
1594 trace_nfs_create_enter(dir, dentry, open_flags);
1595 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1596 trace_nfs_create_exit(dir, dentry, open_flags, error);
1604 EXPORT_SYMBOL_GPL(nfs_create);
1607 * See comments for nfs_proc_create regarding failed operations.
1610 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1615 dfprintk(VFS, "NFS: mknod(%s/%ld), %pd\n",
1616 dir->i_sb->s_id, dir->i_ino, dentry);
1618 if (!new_valid_dev(rdev))
1621 attr.ia_mode = mode;
1622 attr.ia_valid = ATTR_MODE;
1624 trace_nfs_mknod_enter(dir, dentry);
1625 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1626 trace_nfs_mknod_exit(dir, dentry, status);
1634 EXPORT_SYMBOL_GPL(nfs_mknod);
1637 * See comments for nfs_proc_create regarding failed operations.
1639 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1644 dfprintk(VFS, "NFS: mkdir(%s/%ld), %pd\n",
1645 dir->i_sb->s_id, dir->i_ino, dentry);
1647 attr.ia_valid = ATTR_MODE;
1648 attr.ia_mode = mode | S_IFDIR;
1650 trace_nfs_mkdir_enter(dir, dentry);
1651 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1652 trace_nfs_mkdir_exit(dir, dentry, error);
1660 EXPORT_SYMBOL_GPL(nfs_mkdir);
1662 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1664 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1668 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1672 dfprintk(VFS, "NFS: rmdir(%s/%ld), %pd\n",
1673 dir->i_sb->s_id, dir->i_ino, dentry);
1675 trace_nfs_rmdir_enter(dir, dentry);
1676 if (dentry->d_inode) {
1677 nfs_wait_on_sillyrename(dentry);
1678 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1679 /* Ensure the VFS deletes this inode */
1682 clear_nlink(dentry->d_inode);
1685 nfs_dentry_handle_enoent(dentry);
1688 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1689 trace_nfs_rmdir_exit(dir, dentry, error);
1693 EXPORT_SYMBOL_GPL(nfs_rmdir);
1696 * Remove a file after making sure there are no pending writes,
1697 * and after checking that the file has only one user.
1699 * We invalidate the attribute cache and free the inode prior to the operation
1700 * to avoid possible races if the server reuses the inode.
1702 static int nfs_safe_remove(struct dentry *dentry)
1704 struct inode *dir = dentry->d_parent->d_inode;
1705 struct inode *inode = dentry->d_inode;
1708 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1710 /* If the dentry was sillyrenamed, we simply call d_delete() */
1711 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1716 trace_nfs_remove_enter(dir, dentry);
1717 if (inode != NULL) {
1718 NFS_PROTO(inode)->return_delegation(inode);
1719 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1721 nfs_drop_nlink(inode);
1723 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1724 if (error == -ENOENT)
1725 nfs_dentry_handle_enoent(dentry);
1726 trace_nfs_remove_exit(dir, dentry, error);
1731 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1732 * belongs to an active ".nfs..." file and we return -EBUSY.
1734 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1736 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1739 int need_rehash = 0;
1741 dfprintk(VFS, "NFS: unlink(%s/%ld, %pd)\n", dir->i_sb->s_id,
1742 dir->i_ino, dentry);
1744 trace_nfs_unlink_enter(dir, dentry);
1745 spin_lock(&dentry->d_lock);
1746 if (d_count(dentry) > 1) {
1747 spin_unlock(&dentry->d_lock);
1748 /* Start asynchronous writeout of the inode */
1749 write_inode_now(dentry->d_inode, 0);
1750 error = nfs_sillyrename(dir, dentry);
1753 if (!d_unhashed(dentry)) {
1757 spin_unlock(&dentry->d_lock);
1758 error = nfs_safe_remove(dentry);
1759 if (!error || error == -ENOENT) {
1760 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1761 } else if (need_rehash)
1764 trace_nfs_unlink_exit(dir, dentry, error);
1767 EXPORT_SYMBOL_GPL(nfs_unlink);
1770 * To create a symbolic link, most file systems instantiate a new inode,
1771 * add a page to it containing the path, then write it out to the disk
1772 * using prepare_write/commit_write.
1774 * Unfortunately the NFS client can't create the in-core inode first
1775 * because it needs a file handle to create an in-core inode (see
1776 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1777 * symlink request has completed on the server.
1779 * So instead we allocate a raw page, copy the symname into it, then do
1780 * the SYMLINK request with the page as the buffer. If it succeeds, we
1781 * now have a new file handle and can instantiate an in-core NFS inode
1782 * and move the raw page into its mapping.
1784 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1789 unsigned int pathlen = strlen(symname);
1792 dfprintk(VFS, "NFS: symlink(%s/%ld, %pd, %s)\n", dir->i_sb->s_id,
1793 dir->i_ino, dentry, symname);
1795 if (pathlen > PAGE_SIZE)
1796 return -ENAMETOOLONG;
1798 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1799 attr.ia_valid = ATTR_MODE;
1801 page = alloc_page(GFP_HIGHUSER);
1805 kaddr = kmap_atomic(page);
1806 memcpy(kaddr, symname, pathlen);
1807 if (pathlen < PAGE_SIZE)
1808 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1809 kunmap_atomic(kaddr);
1811 trace_nfs_symlink_enter(dir, dentry);
1812 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1813 trace_nfs_symlink_exit(dir, dentry, error);
1815 dfprintk(VFS, "NFS: symlink(%s/%ld, %pd, %s) error %d\n",
1816 dir->i_sb->s_id, dir->i_ino,
1817 dentry, symname, error);
1824 * No big deal if we can't add this page to the page cache here.
1825 * READLINK will get the missing page from the server if needed.
1827 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1829 SetPageUptodate(page);
1836 EXPORT_SYMBOL_GPL(nfs_symlink);
1839 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1841 struct inode *inode = old_dentry->d_inode;
1844 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1845 old_dentry, dentry);
1847 trace_nfs_link_enter(inode, dir, dentry);
1848 NFS_PROTO(inode)->return_delegation(inode);
1851 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1854 d_add(dentry, inode);
1856 trace_nfs_link_exit(inode, dir, dentry, error);
1859 EXPORT_SYMBOL_GPL(nfs_link);
1863 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1864 * different file handle for the same inode after a rename (e.g. when
1865 * moving to a different directory). A fail-safe method to do so would
1866 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1867 * rename the old file using the sillyrename stuff. This way, the original
1868 * file in old_dir will go away when the last process iput()s the inode.
1872 * It actually works quite well. One needs to have the possibility for
1873 * at least one ".nfs..." file in each directory the file ever gets
1874 * moved or linked to which happens automagically with the new
1875 * implementation that only depends on the dcache stuff instead of
1876 * using the inode layer
1878 * Unfortunately, things are a little more complicated than indicated
1879 * above. For a cross-directory move, we want to make sure we can get
1880 * rid of the old inode after the operation. This means there must be
1881 * no pending writes (if it's a file), and the use count must be 1.
1882 * If these conditions are met, we can drop the dentries before doing
1885 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1886 struct inode *new_dir, struct dentry *new_dentry)
1888 struct inode *old_inode = old_dentry->d_inode;
1889 struct inode *new_inode = new_dentry->d_inode;
1890 struct dentry *dentry = NULL, *rehash = NULL;
1893 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1894 old_dentry, new_dentry,
1895 d_count(new_dentry));
1897 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1899 * For non-directories, check whether the target is busy and if so,
1900 * make a copy of the dentry and then do a silly-rename. If the
1901 * silly-rename succeeds, the copied dentry is hashed and becomes
1904 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1906 * To prevent any new references to the target during the
1907 * rename, we unhash the dentry in advance.
1909 if (!d_unhashed(new_dentry)) {
1911 rehash = new_dentry;
1914 if (d_count(new_dentry) > 2) {
1917 /* copy the target dentry's name */
1918 dentry = d_alloc(new_dentry->d_parent,
1919 &new_dentry->d_name);
1923 /* silly-rename the existing target ... */
1924 err = nfs_sillyrename(new_dir, new_dentry);
1928 new_dentry = dentry;
1934 NFS_PROTO(old_inode)->return_delegation(old_inode);
1935 if (new_inode != NULL)
1936 NFS_PROTO(new_inode)->return_delegation(new_inode);
1938 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1939 new_dir, &new_dentry->d_name);
1940 nfs_mark_for_revalidate(old_inode);
1944 trace_nfs_rename_exit(old_dir, old_dentry,
1945 new_dir, new_dentry, error);
1947 if (new_inode != NULL)
1948 nfs_drop_nlink(new_inode);
1949 d_move(old_dentry, new_dentry);
1950 nfs_set_verifier(new_dentry,
1951 nfs_save_change_attribute(new_dir));
1952 } else if (error == -ENOENT)
1953 nfs_dentry_handle_enoent(old_dentry);
1955 /* new dentry created? */
1960 EXPORT_SYMBOL_GPL(nfs_rename);
1962 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1963 static LIST_HEAD(nfs_access_lru_list);
1964 static atomic_long_t nfs_access_nr_entries;
1966 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1968 put_rpccred(entry->cred);
1970 smp_mb__before_atomic_dec();
1971 atomic_long_dec(&nfs_access_nr_entries);
1972 smp_mb__after_atomic_dec();
1975 static void nfs_access_free_list(struct list_head *head)
1977 struct nfs_access_entry *cache;
1979 while (!list_empty(head)) {
1980 cache = list_entry(head->next, struct nfs_access_entry, lru);
1981 list_del(&cache->lru);
1982 nfs_access_free_entry(cache);
1987 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
1990 struct nfs_inode *nfsi, *next;
1991 struct nfs_access_entry *cache;
1992 int nr_to_scan = sc->nr_to_scan;
1993 gfp_t gfp_mask = sc->gfp_mask;
1996 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1999 spin_lock(&nfs_access_lru_lock);
2000 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2001 struct inode *inode;
2003 if (nr_to_scan-- == 0)
2005 inode = &nfsi->vfs_inode;
2006 spin_lock(&inode->i_lock);
2007 if (list_empty(&nfsi->access_cache_entry_lru))
2008 goto remove_lru_entry;
2009 cache = list_entry(nfsi->access_cache_entry_lru.next,
2010 struct nfs_access_entry, lru);
2011 list_move(&cache->lru, &head);
2012 rb_erase(&cache->rb_node, &nfsi->access_cache);
2014 if (!list_empty(&nfsi->access_cache_entry_lru))
2015 list_move_tail(&nfsi->access_cache_inode_lru,
2016 &nfs_access_lru_list);
2019 list_del_init(&nfsi->access_cache_inode_lru);
2020 smp_mb__before_clear_bit();
2021 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2022 smp_mb__after_clear_bit();
2024 spin_unlock(&inode->i_lock);
2026 spin_unlock(&nfs_access_lru_lock);
2027 nfs_access_free_list(&head);
2032 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2034 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2037 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2039 struct rb_root *root_node = &nfsi->access_cache;
2041 struct nfs_access_entry *entry;
2043 /* Unhook entries from the cache */
2044 while ((n = rb_first(root_node)) != NULL) {
2045 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2046 rb_erase(n, root_node);
2047 list_move(&entry->lru, head);
2049 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2052 void nfs_access_zap_cache(struct inode *inode)
2056 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2058 /* Remove from global LRU init */
2059 spin_lock(&nfs_access_lru_lock);
2060 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2061 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2063 spin_lock(&inode->i_lock);
2064 __nfs_access_zap_cache(NFS_I(inode), &head);
2065 spin_unlock(&inode->i_lock);
2066 spin_unlock(&nfs_access_lru_lock);
2067 nfs_access_free_list(&head);
2069 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2071 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2073 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2074 struct nfs_access_entry *entry;
2077 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2079 if (cred < entry->cred)
2081 else if (cred > entry->cred)
2089 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2091 struct nfs_inode *nfsi = NFS_I(inode);
2092 struct nfs_access_entry *cache;
2095 spin_lock(&inode->i_lock);
2096 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2098 cache = nfs_access_search_rbtree(inode, cred);
2101 if (!nfs_have_delegated_attributes(inode) &&
2102 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2104 res->jiffies = cache->jiffies;
2105 res->cred = cache->cred;
2106 res->mask = cache->mask;
2107 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2110 spin_unlock(&inode->i_lock);
2113 rb_erase(&cache->rb_node, &nfsi->access_cache);
2114 list_del(&cache->lru);
2115 spin_unlock(&inode->i_lock);
2116 nfs_access_free_entry(cache);
2119 spin_unlock(&inode->i_lock);
2120 nfs_access_zap_cache(inode);
2124 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2126 struct nfs_inode *nfsi = NFS_I(inode);
2127 struct rb_root *root_node = &nfsi->access_cache;
2128 struct rb_node **p = &root_node->rb_node;
2129 struct rb_node *parent = NULL;
2130 struct nfs_access_entry *entry;
2132 spin_lock(&inode->i_lock);
2133 while (*p != NULL) {
2135 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2137 if (set->cred < entry->cred)
2138 p = &parent->rb_left;
2139 else if (set->cred > entry->cred)
2140 p = &parent->rb_right;
2144 rb_link_node(&set->rb_node, parent, p);
2145 rb_insert_color(&set->rb_node, root_node);
2146 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2147 spin_unlock(&inode->i_lock);
2150 rb_replace_node(parent, &set->rb_node, root_node);
2151 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2152 list_del(&entry->lru);
2153 spin_unlock(&inode->i_lock);
2154 nfs_access_free_entry(entry);
2157 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2159 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2162 RB_CLEAR_NODE(&cache->rb_node);
2163 cache->jiffies = set->jiffies;
2164 cache->cred = get_rpccred(set->cred);
2165 cache->mask = set->mask;
2167 nfs_access_add_rbtree(inode, cache);
2169 /* Update accounting */
2170 smp_mb__before_atomic_inc();
2171 atomic_long_inc(&nfs_access_nr_entries);
2172 smp_mb__after_atomic_inc();
2174 /* Add inode to global LRU list */
2175 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2176 spin_lock(&nfs_access_lru_lock);
2177 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2178 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2179 &nfs_access_lru_list);
2180 spin_unlock(&nfs_access_lru_lock);
2183 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2185 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2188 if (access_result & NFS4_ACCESS_READ)
2189 entry->mask |= MAY_READ;
2191 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2192 entry->mask |= MAY_WRITE;
2193 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2194 entry->mask |= MAY_EXEC;
2196 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2198 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2200 struct nfs_access_entry cache;
2203 trace_nfs_access_enter(inode);
2205 status = nfs_access_get_cached(inode, cred, &cache);
2209 /* Be clever: ask server to check for all possible rights */
2210 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2212 cache.jiffies = jiffies;
2213 status = NFS_PROTO(inode)->access(inode, &cache);
2215 if (status == -ESTALE) {
2216 nfs_zap_caches(inode);
2217 if (!S_ISDIR(inode->i_mode))
2218 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2222 nfs_access_add_cache(inode, &cache);
2224 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2227 trace_nfs_access_exit(inode, status);
2231 static int nfs_open_permission_mask(int openflags)
2235 if (openflags & __FMODE_EXEC) {
2236 /* ONLY check exec rights */
2239 if ((openflags & O_ACCMODE) != O_WRONLY)
2241 if ((openflags & O_ACCMODE) != O_RDONLY)
2248 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2250 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2252 EXPORT_SYMBOL_GPL(nfs_may_open);
2254 int nfs_permission(struct inode *inode, int mask)
2256 struct rpc_cred *cred;
2259 if (mask & MAY_NOT_BLOCK)
2262 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2264 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2266 /* Is this sys_access() ? */
2267 if (mask & (MAY_ACCESS | MAY_CHDIR))
2270 switch (inode->i_mode & S_IFMT) {
2277 * Optimize away all write operations, since the server
2278 * will check permissions when we perform the op.
2280 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2285 if (!NFS_PROTO(inode)->access)
2288 cred = rpc_lookup_cred();
2289 if (!IS_ERR(cred)) {
2290 res = nfs_do_access(inode, cred, mask);
2293 res = PTR_ERR(cred);
2295 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2298 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2299 inode->i_sb->s_id, inode->i_ino, mask, res);
2302 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2304 res = generic_permission(inode, mask);
2307 EXPORT_SYMBOL_GPL(nfs_permission);
2311 * version-control: t
2312 * kept-new-versions: 5