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"
46 /* #define NFS_DEBUG_VERBOSE 1 */
48 static int nfs_opendir(struct inode *, struct file *);
49 static int nfs_closedir(struct inode *, struct file *);
50 static int nfs_readdir(struct file *, struct dir_context *);
51 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
52 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
53 static void nfs_readdir_clear_array(struct page*);
55 const struct file_operations nfs_dir_operations = {
56 .llseek = nfs_llseek_dir,
57 .read = generic_read_dir,
58 .iterate = nfs_readdir,
60 .release = nfs_closedir,
61 .fsync = nfs_fsync_dir,
64 const struct address_space_operations nfs_dir_aops = {
65 .freepage = nfs_readdir_clear_array,
68 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
70 struct nfs_open_dir_context *ctx;
71 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
77 ctx->cred = get_rpccred(cred);
80 return ERR_PTR(-ENOMEM);
83 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
85 put_rpccred(ctx->cred);
93 nfs_opendir(struct inode *inode, struct file *filp)
96 struct nfs_open_dir_context *ctx;
97 struct rpc_cred *cred;
99 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
100 filp->f_path.dentry->d_parent->d_name.name,
101 filp->f_path.dentry->d_name.name);
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 %s/%s contains a readdir loop."
299 "Please contact your server vendor. "
300 "The file: %s has duplicate cookie %llu\n",
301 desc->file->f_dentry->d_parent->d_name.name,
302 desc->file->f_dentry->d_name.name,
303 array->array[i].string.name,
309 ctx->dup_cookie = *desc->dir_cookie;
312 desc->ctx->pos = new_pos;
313 desc->cache_entry_index = i;
317 if (array->eof_index >= 0) {
318 status = -EBADCOOKIE;
319 if (*desc->dir_cookie == array->last_cookie)
327 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
329 struct nfs_cache_array *array;
332 array = nfs_readdir_get_array(desc->page);
334 status = PTR_ERR(array);
338 if (*desc->dir_cookie == 0)
339 status = nfs_readdir_search_for_pos(array, desc);
341 status = nfs_readdir_search_for_cookie(array, desc);
343 if (status == -EAGAIN) {
344 desc->last_cookie = array->last_cookie;
345 desc->current_index += array->size;
348 nfs_readdir_release_array(desc->page);
353 /* Fill a page with xdr information before transferring to the cache page */
355 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
356 struct nfs_entry *entry, struct file *file, struct inode *inode)
358 struct nfs_open_dir_context *ctx = file->private_data;
359 struct rpc_cred *cred = ctx->cred;
360 unsigned long timestamp, gencount;
365 gencount = nfs_inc_attr_generation_counter();
366 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
367 NFS_SERVER(inode)->dtsize, desc->plus);
369 /* We requested READDIRPLUS, but the server doesn't grok it */
370 if (error == -ENOTSUPP && desc->plus) {
371 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
372 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
378 desc->timestamp = timestamp;
379 desc->gencount = gencount;
384 static int xdr_decode(nfs_readdir_descriptor_t *desc,
385 struct nfs_entry *entry, struct xdr_stream *xdr)
389 error = desc->decode(xdr, entry, desc->plus);
392 entry->fattr->time_start = desc->timestamp;
393 entry->fattr->gencount = desc->gencount;
398 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
400 if (dentry->d_inode == NULL)
402 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
410 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
412 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
414 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
422 * This function is called by the lookup code to request the use of
423 * readdirplus to accelerate any future lookups in the same
427 void nfs_advise_use_readdirplus(struct inode *dir)
429 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
433 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
435 struct qstr filename = QSTR_INIT(entry->name, entry->len);
436 struct dentry *dentry;
437 struct dentry *alias;
438 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_refresh_inode(dentry->d_inode, entry->fattr);
455 if (d_invalidate(dentry) != 0)
461 dentry = d_alloc(parent, &filename);
465 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
469 alias = d_materialise_unique(dentry, inode);
473 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
476 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
482 /* Perform conversion from xdr to cache array */
484 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
485 struct page **xdr_pages, struct page *page, unsigned int buflen)
487 struct xdr_stream stream;
489 struct page *scratch;
490 struct nfs_cache_array *array;
491 unsigned int count = 0;
494 scratch = alloc_page(GFP_KERNEL);
498 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
499 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
502 status = xdr_decode(desc, entry, &stream);
504 if (status == -EAGAIN)
512 nfs_prime_dcache(desc->file->f_path.dentry, entry);
514 status = nfs_readdir_add_to_array(entry, page);
517 } while (!entry->eof);
519 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
520 array = nfs_readdir_get_array(page);
521 if (!IS_ERR(array)) {
522 array->eof_index = array->size;
524 nfs_readdir_release_array(page);
526 status = PTR_ERR(array);
534 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
537 for (i = 0; i < npages; i++)
542 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
545 nfs_readdir_free_pagearray(pages, npages);
549 * nfs_readdir_large_page will allocate pages that must be freed with a call
550 * to nfs_readdir_free_large_page
553 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
557 for (i = 0; i < npages; i++) {
558 struct page *page = alloc_page(GFP_KERNEL);
566 nfs_readdir_free_pagearray(pages, i);
571 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
573 struct page *pages[NFS_MAX_READDIR_PAGES];
574 void *pages_ptr = NULL;
575 struct nfs_entry entry;
576 struct file *file = desc->file;
577 struct nfs_cache_array *array;
578 int status = -ENOMEM;
579 unsigned int array_size = ARRAY_SIZE(pages);
581 entry.prev_cookie = 0;
582 entry.cookie = desc->last_cookie;
584 entry.fh = nfs_alloc_fhandle();
585 entry.fattr = nfs_alloc_fattr();
586 entry.server = NFS_SERVER(inode);
587 if (entry.fh == NULL || entry.fattr == NULL)
590 array = nfs_readdir_get_array(page);
592 status = PTR_ERR(array);
595 memset(array, 0, sizeof(struct nfs_cache_array));
596 array->eof_index = -1;
598 status = nfs_readdir_large_page(pages, array_size);
600 goto out_release_array;
603 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
608 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
610 if (status == -ENOSPC)
614 } while (array->eof_index < 0);
616 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
618 nfs_readdir_release_array(page);
620 nfs_free_fattr(entry.fattr);
621 nfs_free_fhandle(entry.fh);
626 * Now we cache directories properly, by converting xdr information
627 * to an array that can be used for lookups later. This results in
628 * fewer cache pages, since we can store more information on each page.
629 * We only need to convert from xdr once so future lookups are much simpler
632 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
634 struct inode *inode = file_inode(desc->file);
637 ret = nfs_readdir_xdr_to_array(desc, page, inode);
640 SetPageUptodate(page);
642 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
643 /* Should never happen */
644 nfs_zap_mapping(inode, inode->i_mapping);
654 void cache_page_release(nfs_readdir_descriptor_t *desc)
656 if (!desc->page->mapping)
657 nfs_readdir_clear_array(desc->page);
658 page_cache_release(desc->page);
663 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
665 return read_cache_page(file_inode(desc->file)->i_mapping,
666 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
670 * Returns 0 if desc->dir_cookie was found on page desc->page_index
673 int find_cache_page(nfs_readdir_descriptor_t *desc)
677 desc->page = get_cache_page(desc);
678 if (IS_ERR(desc->page))
679 return PTR_ERR(desc->page);
681 res = nfs_readdir_search_array(desc);
683 cache_page_release(desc);
687 /* Search for desc->dir_cookie from the beginning of the page cache */
689 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
693 if (desc->page_index == 0) {
694 desc->current_index = 0;
695 desc->last_cookie = 0;
698 res = find_cache_page(desc);
699 } while (res == -EAGAIN);
704 * Once we've found the start of the dirent within a page: fill 'er up...
707 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
709 struct file *file = desc->file;
712 struct nfs_cache_array *array = NULL;
713 struct nfs_open_dir_context *ctx = file->private_data;
715 array = nfs_readdir_get_array(desc->page);
717 res = PTR_ERR(array);
721 for (i = desc->cache_entry_index; i < array->size; i++) {
722 struct nfs_cache_array_entry *ent;
724 ent = &array->array[i];
725 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
726 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
731 if (i < (array->size-1))
732 *desc->dir_cookie = array->array[i+1].cookie;
734 *desc->dir_cookie = array->last_cookie;
738 if (array->eof_index >= 0)
741 nfs_readdir_release_array(desc->page);
743 cache_page_release(desc);
744 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
745 (unsigned long long)*desc->dir_cookie, res);
750 * If we cannot find a cookie in our cache, we suspect that this is
751 * because it points to a deleted file, so we ask the server to return
752 * whatever it thinks is the next entry. We then feed this to filldir.
753 * If all goes well, we should then be able to find our way round the
754 * cache on the next call to readdir_search_pagecache();
756 * NOTE: we cannot add the anonymous page to the pagecache because
757 * the data it contains might not be page aligned. Besides,
758 * we should already have a complete representation of the
759 * directory in the page cache by the time we get here.
762 int uncached_readdir(nfs_readdir_descriptor_t *desc)
764 struct page *page = NULL;
766 struct inode *inode = file_inode(desc->file);
767 struct nfs_open_dir_context *ctx = desc->file->private_data;
769 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
770 (unsigned long long)*desc->dir_cookie);
772 page = alloc_page(GFP_HIGHUSER);
778 desc->page_index = 0;
779 desc->last_cookie = *desc->dir_cookie;
783 status = nfs_readdir_xdr_to_array(desc, page, inode);
787 status = nfs_do_filldir(desc);
790 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
794 cache_page_release(desc);
798 /* The file offset position represents the dirent entry number. A
799 last cookie cache takes care of the common case of reading the
802 static int nfs_readdir(struct file *file, struct dir_context *ctx)
804 struct dentry *dentry = file->f_path.dentry;
805 struct inode *inode = dentry->d_inode;
806 nfs_readdir_descriptor_t my_desc,
808 struct nfs_open_dir_context *dir_ctx = file->private_data;
811 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
812 dentry->d_parent->d_name.name, dentry->d_name.name,
813 (long long)ctx->pos);
814 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
817 * ctx->pos points to the dirent entry number.
818 * *desc->dir_cookie has the cookie for the next entry. We have
819 * to either find the entry with the appropriate number or
820 * revalidate the cookie.
822 memset(desc, 0, sizeof(*desc));
826 desc->dir_cookie = &dir_ctx->dir_cookie;
827 desc->decode = NFS_PROTO(inode)->decode_dirent;
828 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
830 nfs_block_sillyrename(dentry);
831 res = nfs_revalidate_mapping(inode, file->f_mapping);
836 res = readdir_search_pagecache(desc);
838 if (res == -EBADCOOKIE) {
840 /* This means either end of directory */
841 if (*desc->dir_cookie && desc->eof == 0) {
842 /* Or that the server has 'lost' a cookie */
843 res = uncached_readdir(desc);
849 if (res == -ETOOSMALL && desc->plus) {
850 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
851 nfs_zap_caches(inode);
852 desc->page_index = 0;
860 res = nfs_do_filldir(desc);
863 } while (!desc->eof);
865 nfs_unblock_sillyrename(dentry);
868 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
869 dentry->d_parent->d_name.name, dentry->d_name.name,
874 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
876 struct dentry *dentry = filp->f_path.dentry;
877 struct inode *inode = dentry->d_inode;
878 struct nfs_open_dir_context *dir_ctx = filp->private_data;
880 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
881 dentry->d_parent->d_name.name,
885 mutex_lock(&inode->i_mutex);
888 offset += filp->f_pos;
896 if (offset != filp->f_pos) {
897 filp->f_pos = offset;
898 dir_ctx->dir_cookie = 0;
902 mutex_unlock(&inode->i_mutex);
907 * All directory operations under NFS are synchronous, so fsync()
908 * is a dummy operation.
910 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
913 struct dentry *dentry = filp->f_path.dentry;
914 struct inode *inode = dentry->d_inode;
916 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
917 dentry->d_parent->d_name.name, dentry->d_name.name,
920 mutex_lock(&inode->i_mutex);
921 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
922 mutex_unlock(&inode->i_mutex);
927 * nfs_force_lookup_revalidate - Mark the directory as having changed
928 * @dir - pointer to directory inode
930 * This forces the revalidation code in nfs_lookup_revalidate() to do a
931 * full lookup on all child dentries of 'dir' whenever a change occurs
932 * on the server that might have invalidated our dcache.
934 * The caller should be holding dir->i_lock
936 void nfs_force_lookup_revalidate(struct inode *dir)
938 NFS_I(dir)->cache_change_attribute++;
940 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
943 * A check for whether or not the parent directory has changed.
944 * In the case it has, we assume that the dentries are untrustworthy
945 * and may need to be looked up again.
947 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
951 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
953 if (!nfs_verify_change_attribute(dir, dentry->d_time))
955 /* Revalidate nfsi->cache_change_attribute before we declare a match */
956 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
958 if (!nfs_verify_change_attribute(dir, dentry->d_time))
964 * Use intent information to check whether or not we're going to do
965 * an O_EXCL create using this path component.
967 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
969 if (NFS_PROTO(dir)->version == 2)
971 return flags & LOOKUP_EXCL;
975 * Inode and filehandle revalidation for lookups.
977 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
978 * or if the intent information indicates that we're about to open this
979 * particular file and the "nocto" mount flag is not set.
983 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
985 struct nfs_server *server = NFS_SERVER(inode);
988 if (IS_AUTOMOUNT(inode))
990 /* VFS wants an on-the-wire revalidation */
991 if (flags & LOOKUP_REVAL)
993 /* This is an open(2) */
994 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
995 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
998 return (inode->i_nlink == 0) ? -ENOENT : 0;
1000 ret = __nfs_revalidate_inode(server, inode);
1007 * We judge how long we want to trust negative
1008 * dentries by looking at the parent inode mtime.
1010 * If parent mtime has changed, we revalidate, else we wait for a
1011 * period corresponding to the parent's attribute cache timeout value.
1014 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1017 /* Don't revalidate a negative dentry if we're creating a new file */
1018 if (flags & LOOKUP_CREATE)
1020 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1022 return !nfs_check_verifier(dir, dentry);
1026 * This is called every time the dcache has a lookup hit,
1027 * and we should check whether we can really trust that
1030 * NOTE! The hit can be a negative hit too, don't assume
1033 * If the parent directory is seen to have changed, we throw out the
1034 * cached dentry and do a new lookup.
1036 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1039 struct inode *inode;
1040 struct dentry *parent;
1041 struct nfs_fh *fhandle = NULL;
1042 struct nfs_fattr *fattr = NULL;
1045 if (flags & LOOKUP_RCU)
1048 parent = dget_parent(dentry);
1049 dir = parent->d_inode;
1050 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1051 inode = dentry->d_inode;
1054 if (nfs_neg_need_reval(dir, dentry, flags))
1056 goto out_valid_noent;
1059 if (is_bad_inode(inode)) {
1060 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1061 __func__, dentry->d_parent->d_name.name,
1062 dentry->d_name.name);
1066 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1067 goto out_set_verifier;
1069 /* Force a full look up iff the parent directory has changed */
1070 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1071 if (nfs_lookup_verify_inode(inode, flags))
1072 goto out_zap_parent;
1076 if (NFS_STALE(inode))
1080 fhandle = nfs_alloc_fhandle();
1081 fattr = nfs_alloc_fattr();
1082 if (fhandle == NULL || fattr == NULL)
1085 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1088 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1090 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1093 nfs_free_fattr(fattr);
1094 nfs_free_fhandle(fhandle);
1096 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1098 /* Success: notify readdir to use READDIRPLUS */
1099 nfs_advise_use_readdirplus(dir);
1102 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1103 __func__, dentry->d_parent->d_name.name,
1104 dentry->d_name.name);
1107 nfs_zap_caches(dir);
1109 nfs_free_fattr(fattr);
1110 nfs_free_fhandle(fhandle);
1111 nfs_mark_for_revalidate(dir);
1112 if (inode && S_ISDIR(inode->i_mode)) {
1113 /* Purge readdir caches. */
1114 nfs_zap_caches(inode);
1115 /* If we have submounts, don't unhash ! */
1116 if (have_submounts(dentry))
1118 if (dentry->d_flags & DCACHE_DISCONNECTED)
1120 shrink_dcache_parent(dentry);
1124 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1125 __func__, dentry->d_parent->d_name.name,
1126 dentry->d_name.name);
1129 nfs_free_fattr(fattr);
1130 nfs_free_fhandle(fhandle);
1132 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1133 __func__, dentry->d_parent->d_name.name,
1134 dentry->d_name.name, error);
1139 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1140 * when we don't really care about the dentry name. This is called when a
1141 * pathwalk ends on a dentry that was not found via a normal lookup in the
1142 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1144 * In this situation, we just want to verify that the inode itself is OK
1145 * since the dentry might have changed on the server.
1147 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1150 struct inode *inode = dentry->d_inode;
1153 * I believe we can only get a negative dentry here in the case of a
1154 * procfs-style symlink. Just assume it's correct for now, but we may
1155 * eventually need to do something more here.
1158 dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1159 __func__, dentry->d_parent->d_name.name,
1160 dentry->d_name.name);
1164 if (is_bad_inode(inode)) {
1165 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1166 __func__, dentry->d_parent->d_name.name,
1167 dentry->d_name.name);
1171 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1172 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1173 __func__, inode->i_ino, error ? "invalid" : "valid");
1178 * This is called from dput() when d_count is going to 0.
1180 static int nfs_dentry_delete(const struct dentry *dentry)
1182 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1183 dentry->d_parent->d_name.name, dentry->d_name.name,
1186 /* Unhash any dentry with a stale inode */
1187 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1190 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1191 /* Unhash it, so that ->d_iput() would be called */
1194 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1195 /* Unhash it, so that ancestors of killed async unlink
1196 * files will be cleaned up during umount */
1203 /* Ensure that we revalidate inode->i_nlink */
1204 static void nfs_drop_nlink(struct inode *inode)
1206 spin_lock(&inode->i_lock);
1207 /* drop the inode if we're reasonably sure this is the last link */
1208 if (inode->i_nlink == 1)
1210 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1211 spin_unlock(&inode->i_lock);
1215 * Called when the dentry loses inode.
1216 * We use it to clean up silly-renamed files.
1218 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1220 if (S_ISDIR(inode->i_mode))
1221 /* drop any readdir cache as it could easily be old */
1222 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1224 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1225 nfs_complete_unlink(dentry, inode);
1226 nfs_drop_nlink(inode);
1231 static void nfs_d_release(struct dentry *dentry)
1233 /* free cached devname value, if it survived that far */
1234 if (unlikely(dentry->d_fsdata)) {
1235 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1238 kfree(dentry->d_fsdata);
1242 const struct dentry_operations nfs_dentry_operations = {
1243 .d_revalidate = nfs_lookup_revalidate,
1244 .d_weak_revalidate = nfs_weak_revalidate,
1245 .d_delete = nfs_dentry_delete,
1246 .d_iput = nfs_dentry_iput,
1247 .d_automount = nfs_d_automount,
1248 .d_release = nfs_d_release,
1250 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1252 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1255 struct dentry *parent;
1256 struct inode *inode = NULL;
1257 struct nfs_fh *fhandle = NULL;
1258 struct nfs_fattr *fattr = NULL;
1261 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1262 dentry->d_parent->d_name.name, dentry->d_name.name);
1263 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1265 res = ERR_PTR(-ENAMETOOLONG);
1266 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1270 * If we're doing an exclusive create, optimize away the lookup
1271 * but don't hash the dentry.
1273 if (nfs_is_exclusive_create(dir, flags)) {
1274 d_instantiate(dentry, NULL);
1279 res = ERR_PTR(-ENOMEM);
1280 fhandle = nfs_alloc_fhandle();
1281 fattr = nfs_alloc_fattr();
1282 if (fhandle == NULL || fattr == NULL)
1285 parent = dentry->d_parent;
1286 /* Protect against concurrent sillydeletes */
1287 nfs_block_sillyrename(parent);
1288 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1289 if (error == -ENOENT)
1292 res = ERR_PTR(error);
1293 goto out_unblock_sillyrename;
1295 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1296 res = ERR_CAST(inode);
1298 goto out_unblock_sillyrename;
1300 /* Success: notify readdir to use READDIRPLUS */
1301 nfs_advise_use_readdirplus(dir);
1304 res = d_materialise_unique(dentry, inode);
1307 goto out_unblock_sillyrename;
1310 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1311 out_unblock_sillyrename:
1312 nfs_unblock_sillyrename(parent);
1314 nfs_free_fattr(fattr);
1315 nfs_free_fhandle(fhandle);
1318 EXPORT_SYMBOL_GPL(nfs_lookup);
1320 #if IS_ENABLED(CONFIG_NFS_V4)
1321 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1323 const struct dentry_operations nfs4_dentry_operations = {
1324 .d_revalidate = nfs4_lookup_revalidate,
1325 .d_delete = nfs_dentry_delete,
1326 .d_iput = nfs_dentry_iput,
1327 .d_automount = nfs_d_automount,
1328 .d_release = nfs_d_release,
1330 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1332 static fmode_t flags_to_mode(int flags)
1334 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1335 if ((flags & O_ACCMODE) != O_WRONLY)
1337 if ((flags & O_ACCMODE) != O_RDONLY)
1342 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1344 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1347 static int do_open(struct inode *inode, struct file *filp)
1349 nfs_fscache_set_inode_cookie(inode, filp);
1353 static int nfs_finish_open(struct nfs_open_context *ctx,
1354 struct dentry *dentry,
1355 struct file *file, unsigned open_flags,
1360 if (ctx->dentry != dentry) {
1362 ctx->dentry = dget(dentry);
1365 /* If the open_intent is for execute, we have an extra check to make */
1366 if (ctx->mode & FMODE_EXEC) {
1367 err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1372 err = finish_open(file, dentry, do_open, opened);
1375 nfs_file_set_open_context(file, ctx);
1378 put_nfs_open_context(ctx);
1382 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1383 struct file *file, unsigned open_flags,
1384 umode_t mode, int *opened)
1386 struct nfs_open_context *ctx;
1388 struct iattr attr = { .ia_valid = ATTR_OPEN };
1389 struct inode *inode;
1392 /* Expect a negative dentry */
1393 BUG_ON(dentry->d_inode);
1395 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1396 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1398 /* NFS only supports OPEN on regular files */
1399 if ((open_flags & O_DIRECTORY)) {
1400 if (!d_unhashed(dentry)) {
1402 * Hashed negative dentry with O_DIRECTORY: dentry was
1403 * revalidated and is fine, no need to perform lookup
1411 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1412 return -ENAMETOOLONG;
1414 if (open_flags & O_CREAT) {
1415 attr.ia_valid |= ATTR_MODE;
1416 attr.ia_mode = mode & ~current_umask();
1418 if (open_flags & O_TRUNC) {
1419 attr.ia_valid |= ATTR_SIZE;
1423 ctx = create_nfs_open_context(dentry, open_flags);
1428 nfs_block_sillyrename(dentry->d_parent);
1429 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1431 if (IS_ERR(inode)) {
1432 nfs_unblock_sillyrename(dentry->d_parent);
1433 put_nfs_open_context(ctx);
1434 err = PTR_ERR(inode);
1437 d_add(dentry, NULL);
1443 if (!(open_flags & O_NOFOLLOW))
1452 res = d_add_unique(dentry, inode);
1456 nfs_unblock_sillyrename(dentry->d_parent);
1457 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1459 err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1466 res = nfs_lookup(dir, dentry, 0);
1471 return finish_no_open(file, res);
1473 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1475 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1477 struct dentry *parent = NULL;
1478 struct inode *inode;
1482 if (flags & LOOKUP_RCU)
1485 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1487 if (d_mountpoint(dentry))
1489 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1492 inode = dentry->d_inode;
1493 parent = dget_parent(dentry);
1494 dir = parent->d_inode;
1496 /* We can't create new files in nfs_open_revalidate(), so we
1497 * optimize away revalidation of negative dentries.
1499 if (inode == NULL) {
1500 if (!nfs_neg_need_reval(dir, dentry, flags))
1505 /* NFS only supports OPEN on regular files */
1506 if (!S_ISREG(inode->i_mode))
1508 /* We cannot do exclusive creation on a positive dentry */
1509 if (flags & LOOKUP_EXCL)
1512 /* Let f_op->open() actually open (and revalidate) the file */
1522 return nfs_lookup_revalidate(dentry, flags);
1525 #endif /* CONFIG_NFSV4 */
1528 * Code common to create, mkdir, and mknod.
1530 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1531 struct nfs_fattr *fattr)
1533 struct dentry *parent = dget_parent(dentry);
1534 struct inode *dir = parent->d_inode;
1535 struct inode *inode;
1536 int error = -EACCES;
1540 /* We may have been initialized further down */
1541 if (dentry->d_inode)
1543 if (fhandle->size == 0) {
1544 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1548 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1549 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1550 struct nfs_server *server = NFS_SB(dentry->d_sb);
1551 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1555 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1556 error = PTR_ERR(inode);
1559 d_add(dentry, inode);
1564 nfs_mark_for_revalidate(dir);
1568 EXPORT_SYMBOL_GPL(nfs_instantiate);
1571 * Following a failed create operation, we drop the dentry rather
1572 * than retain a negative dentry. This avoids a problem in the event
1573 * that the operation succeeded on the server, but an error in the
1574 * reply path made it appear to have failed.
1576 int nfs_create(struct inode *dir, struct dentry *dentry,
1577 umode_t mode, bool excl)
1580 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1583 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1584 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1586 attr.ia_mode = mode;
1587 attr.ia_valid = ATTR_MODE;
1589 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1597 EXPORT_SYMBOL_GPL(nfs_create);
1600 * See comments for nfs_proc_create regarding failed operations.
1603 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1608 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1609 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1611 if (!new_valid_dev(rdev))
1614 attr.ia_mode = mode;
1615 attr.ia_valid = ATTR_MODE;
1617 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1625 EXPORT_SYMBOL_GPL(nfs_mknod);
1628 * See comments for nfs_proc_create regarding failed operations.
1630 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1635 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1636 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1638 attr.ia_valid = ATTR_MODE;
1639 attr.ia_mode = mode | S_IFDIR;
1641 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1649 EXPORT_SYMBOL_GPL(nfs_mkdir);
1651 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1653 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1657 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1661 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1662 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1664 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1665 /* Ensure the VFS deletes this inode */
1666 if (error == 0 && dentry->d_inode != NULL)
1667 clear_nlink(dentry->d_inode);
1668 else if (error == -ENOENT)
1669 nfs_dentry_handle_enoent(dentry);
1673 EXPORT_SYMBOL_GPL(nfs_rmdir);
1676 * Remove a file after making sure there are no pending writes,
1677 * and after checking that the file has only one user.
1679 * We invalidate the attribute cache and free the inode prior to the operation
1680 * to avoid possible races if the server reuses the inode.
1682 static int nfs_safe_remove(struct dentry *dentry)
1684 struct inode *dir = dentry->d_parent->d_inode;
1685 struct inode *inode = dentry->d_inode;
1688 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1689 dentry->d_parent->d_name.name, dentry->d_name.name);
1691 /* If the dentry was sillyrenamed, we simply call d_delete() */
1692 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1697 if (inode != NULL) {
1698 NFS_PROTO(inode)->return_delegation(inode);
1699 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1701 nfs_drop_nlink(inode);
1703 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1704 if (error == -ENOENT)
1705 nfs_dentry_handle_enoent(dentry);
1710 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1711 * belongs to an active ".nfs..." file and we return -EBUSY.
1713 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1715 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1718 int need_rehash = 0;
1720 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1721 dir->i_ino, dentry->d_name.name);
1723 spin_lock(&dentry->d_lock);
1724 if (d_count(dentry) > 1) {
1725 spin_unlock(&dentry->d_lock);
1726 /* Start asynchronous writeout of the inode */
1727 write_inode_now(dentry->d_inode, 0);
1728 error = nfs_sillyrename(dir, dentry);
1731 if (!d_unhashed(dentry)) {
1735 spin_unlock(&dentry->d_lock);
1736 error = nfs_safe_remove(dentry);
1737 if (!error || error == -ENOENT) {
1738 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1739 } else if (need_rehash)
1743 EXPORT_SYMBOL_GPL(nfs_unlink);
1746 * To create a symbolic link, most file systems instantiate a new inode,
1747 * add a page to it containing the path, then write it out to the disk
1748 * using prepare_write/commit_write.
1750 * Unfortunately the NFS client can't create the in-core inode first
1751 * because it needs a file handle to create an in-core inode (see
1752 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1753 * symlink request has completed on the server.
1755 * So instead we allocate a raw page, copy the symname into it, then do
1756 * the SYMLINK request with the page as the buffer. If it succeeds, we
1757 * now have a new file handle and can instantiate an in-core NFS inode
1758 * and move the raw page into its mapping.
1760 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
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 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1803 SetPageUptodate(page);
1810 EXPORT_SYMBOL_GPL(nfs_symlink);
1813 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1815 struct inode *inode = old_dentry->d_inode;
1818 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1819 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1820 dentry->d_parent->d_name.name, dentry->d_name.name);
1822 NFS_PROTO(inode)->return_delegation(inode);
1825 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1828 d_add(dentry, inode);
1832 EXPORT_SYMBOL_GPL(nfs_link);
1836 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1837 * different file handle for the same inode after a rename (e.g. when
1838 * moving to a different directory). A fail-safe method to do so would
1839 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1840 * rename the old file using the sillyrename stuff. This way, the original
1841 * file in old_dir will go away when the last process iput()s the inode.
1845 * It actually works quite well. One needs to have the possibility for
1846 * at least one ".nfs..." file in each directory the file ever gets
1847 * moved or linked to which happens automagically with the new
1848 * implementation that only depends on the dcache stuff instead of
1849 * using the inode layer
1851 * Unfortunately, things are a little more complicated than indicated
1852 * above. For a cross-directory move, we want to make sure we can get
1853 * rid of the old inode after the operation. This means there must be
1854 * no pending writes (if it's a file), and the use count must be 1.
1855 * If these conditions are met, we can drop the dentries before doing
1858 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1859 struct inode *new_dir, struct dentry *new_dentry)
1861 struct inode *old_inode = old_dentry->d_inode;
1862 struct inode *new_inode = new_dentry->d_inode;
1863 struct dentry *dentry = NULL, *rehash = NULL;
1866 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1867 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1868 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1869 d_count(new_dentry));
1872 * For non-directories, check whether the target is busy and if so,
1873 * make a copy of the dentry and then do a silly-rename. If the
1874 * silly-rename succeeds, the copied dentry is hashed and becomes
1877 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1879 * To prevent any new references to the target during the
1880 * rename, we unhash the dentry in advance.
1882 if (!d_unhashed(new_dentry)) {
1884 rehash = new_dentry;
1887 if (d_count(new_dentry) > 2) {
1890 /* copy the target dentry's name */
1891 dentry = d_alloc(new_dentry->d_parent,
1892 &new_dentry->d_name);
1896 /* silly-rename the existing target ... */
1897 err = nfs_sillyrename(new_dir, new_dentry);
1901 new_dentry = dentry;
1907 NFS_PROTO(old_inode)->return_delegation(old_inode);
1908 if (new_inode != NULL)
1909 NFS_PROTO(new_inode)->return_delegation(new_inode);
1911 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1912 new_dir, &new_dentry->d_name);
1913 nfs_mark_for_revalidate(old_inode);
1918 if (new_inode != NULL)
1919 nfs_drop_nlink(new_inode);
1920 d_move(old_dentry, new_dentry);
1921 nfs_set_verifier(new_dentry,
1922 nfs_save_change_attribute(new_dir));
1923 } else if (error == -ENOENT)
1924 nfs_dentry_handle_enoent(old_dentry);
1926 /* new dentry created? */
1931 EXPORT_SYMBOL_GPL(nfs_rename);
1933 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1934 static LIST_HEAD(nfs_access_lru_list);
1935 static atomic_long_t nfs_access_nr_entries;
1937 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1939 put_rpccred(entry->cred);
1941 smp_mb__before_atomic_dec();
1942 atomic_long_dec(&nfs_access_nr_entries);
1943 smp_mb__after_atomic_dec();
1946 static void nfs_access_free_list(struct list_head *head)
1948 struct nfs_access_entry *cache;
1950 while (!list_empty(head)) {
1951 cache = list_entry(head->next, struct nfs_access_entry, lru);
1952 list_del(&cache->lru);
1953 nfs_access_free_entry(cache);
1957 int nfs_access_cache_shrinker(struct shrinker *shrink,
1958 struct shrink_control *sc)
1961 struct nfs_inode *nfsi, *next;
1962 struct nfs_access_entry *cache;
1963 int nr_to_scan = sc->nr_to_scan;
1964 gfp_t gfp_mask = sc->gfp_mask;
1966 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1967 return (nr_to_scan == 0) ? 0 : -1;
1969 spin_lock(&nfs_access_lru_lock);
1970 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1971 struct inode *inode;
1973 if (nr_to_scan-- == 0)
1975 inode = &nfsi->vfs_inode;
1976 spin_lock(&inode->i_lock);
1977 if (list_empty(&nfsi->access_cache_entry_lru))
1978 goto remove_lru_entry;
1979 cache = list_entry(nfsi->access_cache_entry_lru.next,
1980 struct nfs_access_entry, lru);
1981 list_move(&cache->lru, &head);
1982 rb_erase(&cache->rb_node, &nfsi->access_cache);
1983 if (!list_empty(&nfsi->access_cache_entry_lru))
1984 list_move_tail(&nfsi->access_cache_inode_lru,
1985 &nfs_access_lru_list);
1988 list_del_init(&nfsi->access_cache_inode_lru);
1989 smp_mb__before_clear_bit();
1990 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1991 smp_mb__after_clear_bit();
1993 spin_unlock(&inode->i_lock);
1995 spin_unlock(&nfs_access_lru_lock);
1996 nfs_access_free_list(&head);
1997 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2000 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2002 struct rb_root *root_node = &nfsi->access_cache;
2004 struct nfs_access_entry *entry;
2006 /* Unhook entries from the cache */
2007 while ((n = rb_first(root_node)) != NULL) {
2008 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2009 rb_erase(n, root_node);
2010 list_move(&entry->lru, head);
2012 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2015 void nfs_access_zap_cache(struct inode *inode)
2019 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2021 /* Remove from global LRU init */
2022 spin_lock(&nfs_access_lru_lock);
2023 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2024 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2026 spin_lock(&inode->i_lock);
2027 __nfs_access_zap_cache(NFS_I(inode), &head);
2028 spin_unlock(&inode->i_lock);
2029 spin_unlock(&nfs_access_lru_lock);
2030 nfs_access_free_list(&head);
2032 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2034 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2036 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2037 struct nfs_access_entry *entry;
2040 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2042 if (cred < entry->cred)
2044 else if (cred > entry->cred)
2052 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2054 struct nfs_inode *nfsi = NFS_I(inode);
2055 struct nfs_access_entry *cache;
2058 spin_lock(&inode->i_lock);
2059 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2061 cache = nfs_access_search_rbtree(inode, cred);
2064 if (!nfs_have_delegated_attributes(inode) &&
2065 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2067 res->jiffies = cache->jiffies;
2068 res->cred = cache->cred;
2069 res->mask = cache->mask;
2070 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2073 spin_unlock(&inode->i_lock);
2076 rb_erase(&cache->rb_node, &nfsi->access_cache);
2077 list_del(&cache->lru);
2078 spin_unlock(&inode->i_lock);
2079 nfs_access_free_entry(cache);
2082 spin_unlock(&inode->i_lock);
2083 nfs_access_zap_cache(inode);
2087 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2089 struct nfs_inode *nfsi = NFS_I(inode);
2090 struct rb_root *root_node = &nfsi->access_cache;
2091 struct rb_node **p = &root_node->rb_node;
2092 struct rb_node *parent = NULL;
2093 struct nfs_access_entry *entry;
2095 spin_lock(&inode->i_lock);
2096 while (*p != NULL) {
2098 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2100 if (set->cred < entry->cred)
2101 p = &parent->rb_left;
2102 else if (set->cred > entry->cred)
2103 p = &parent->rb_right;
2107 rb_link_node(&set->rb_node, parent, p);
2108 rb_insert_color(&set->rb_node, root_node);
2109 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2110 spin_unlock(&inode->i_lock);
2113 rb_replace_node(parent, &set->rb_node, root_node);
2114 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2115 list_del(&entry->lru);
2116 spin_unlock(&inode->i_lock);
2117 nfs_access_free_entry(entry);
2120 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2122 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2125 RB_CLEAR_NODE(&cache->rb_node);
2126 cache->jiffies = set->jiffies;
2127 cache->cred = get_rpccred(set->cred);
2128 cache->mask = set->mask;
2130 nfs_access_add_rbtree(inode, cache);
2132 /* Update accounting */
2133 smp_mb__before_atomic_inc();
2134 atomic_long_inc(&nfs_access_nr_entries);
2135 smp_mb__after_atomic_inc();
2137 /* Add inode to global LRU list */
2138 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2139 spin_lock(&nfs_access_lru_lock);
2140 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2141 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2142 &nfs_access_lru_list);
2143 spin_unlock(&nfs_access_lru_lock);
2146 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2148 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2151 if (access_result & NFS4_ACCESS_READ)
2152 entry->mask |= MAY_READ;
2154 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2155 entry->mask |= MAY_WRITE;
2156 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2157 entry->mask |= MAY_EXEC;
2159 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2161 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2163 struct nfs_access_entry cache;
2166 status = nfs_access_get_cached(inode, cred, &cache);
2170 /* Be clever: ask server to check for all possible rights */
2171 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2173 cache.jiffies = jiffies;
2174 status = NFS_PROTO(inode)->access(inode, &cache);
2176 if (status == -ESTALE) {
2177 nfs_zap_caches(inode);
2178 if (!S_ISDIR(inode->i_mode))
2179 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2183 nfs_access_add_cache(inode, &cache);
2185 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2190 static int nfs_open_permission_mask(int openflags)
2194 if (openflags & __FMODE_EXEC) {
2195 /* ONLY check exec rights */
2198 if ((openflags & O_ACCMODE) != O_WRONLY)
2200 if ((openflags & O_ACCMODE) != O_RDONLY)
2207 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2209 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2211 EXPORT_SYMBOL_GPL(nfs_may_open);
2213 int nfs_permission(struct inode *inode, int mask)
2215 struct rpc_cred *cred;
2218 if (mask & MAY_NOT_BLOCK)
2221 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2223 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2225 /* Is this sys_access() ? */
2226 if (mask & (MAY_ACCESS | MAY_CHDIR))
2229 switch (inode->i_mode & S_IFMT) {
2233 /* NFSv4 has atomic_open... */
2234 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2235 && (mask & MAY_OPEN)
2236 && !(mask & MAY_EXEC))
2241 * Optimize away all write operations, since the server
2242 * will check permissions when we perform the op.
2244 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2249 if (!NFS_PROTO(inode)->access)
2252 cred = rpc_lookup_cred();
2253 if (!IS_ERR(cred)) {
2254 res = nfs_do_access(inode, cred, mask);
2257 res = PTR_ERR(cred);
2259 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2262 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2263 inode->i_sb->s_id, inode->i_ino, mask, res);
2266 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2268 res = generic_permission(inode, mask);
2271 EXPORT_SYMBOL_GPL(nfs_permission);
2275 * version-control: t
2276 * kept-new-versions: 5