4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/vmalloc.h>
37 #include <linux/kmemleak.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_readdir(struct file *, void *, filldir_t);
48 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
49 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
50 static int nfs_mkdir(struct inode *, struct dentry *, int);
51 static int nfs_rmdir(struct inode *, struct dentry *);
52 static int nfs_unlink(struct inode *, struct dentry *);
53 static int nfs_symlink(struct inode *, struct dentry *, const char *);
54 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
55 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
56 static int nfs_rename(struct inode *, struct dentry *,
57 struct inode *, struct dentry *);
58 static int nfs_fsync_dir(struct file *, int);
59 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
60 static int nfs_readdir_clear_array(struct page*, gfp_t);
62 const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .readdir = nfs_readdir,
67 .release = nfs_release,
68 .fsync = nfs_fsync_dir,
71 const struct inode_operations nfs_dir_inode_operations = {
76 .symlink = nfs_symlink,
81 .permission = nfs_permission,
82 .getattr = nfs_getattr,
83 .setattr = nfs_setattr,
86 const struct address_space_operations nfs_dir_addr_space_ops = {
87 .releasepage = nfs_readdir_clear_array,
91 const struct inode_operations nfs3_dir_inode_operations = {
96 .symlink = nfs_symlink,
100 .rename = nfs_rename,
101 .permission = nfs_permission,
102 .getattr = nfs_getattr,
103 .setattr = nfs_setattr,
104 .listxattr = nfs3_listxattr,
105 .getxattr = nfs3_getxattr,
106 .setxattr = nfs3_setxattr,
107 .removexattr = nfs3_removexattr,
109 #endif /* CONFIG_NFS_V3 */
113 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
114 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
115 const struct inode_operations nfs4_dir_inode_operations = {
116 .create = nfs_open_create,
117 .lookup = nfs_atomic_lookup,
119 .unlink = nfs_unlink,
120 .symlink = nfs_symlink,
124 .rename = nfs_rename,
125 .permission = nfs_permission,
126 .getattr = nfs_getattr,
127 .setattr = nfs_setattr,
128 .getxattr = nfs4_getxattr,
129 .setxattr = nfs4_setxattr,
130 .listxattr = nfs4_listxattr,
133 #endif /* CONFIG_NFS_V4 */
139 nfs_opendir(struct inode *inode, struct file *filp)
143 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
144 filp->f_path.dentry->d_parent->d_name.name,
145 filp->f_path.dentry->d_name.name);
147 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
149 /* Call generic open code in order to cache credentials */
150 res = nfs_open(inode, filp);
151 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
152 /* This is a mountpoint, so d_revalidate will never
153 * have been called, so we need to refresh the
154 * inode (for close-open consistency) ourselves.
156 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
161 struct nfs_cache_array_entry {
165 unsigned char d_type;
168 struct nfs_cache_array {
172 struct nfs_cache_array_entry array[0];
175 typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int);
179 unsigned long page_index;
181 loff_t current_index;
182 decode_dirent_t decode;
184 unsigned long timestamp;
185 unsigned long gencount;
186 unsigned int cache_entry_index;
189 } nfs_readdir_descriptor_t;
192 * The caller is responsible for calling nfs_readdir_release_array(page)
195 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
199 return ERR_PTR(-EIO);
202 return ERR_PTR(-ENOMEM);
207 void nfs_readdir_release_array(struct page *page)
213 * we are freeing strings created by nfs_add_to_readdir_array()
216 int nfs_readdir_clear_array(struct page *page, gfp_t mask)
218 struct nfs_cache_array *array = nfs_readdir_get_array(page);
222 return PTR_ERR(array);
223 for (i = 0; i < array->size; i++)
224 kfree(array->array[i].string.name);
225 nfs_readdir_release_array(page);
230 * the caller is responsible for freeing qstr.name
231 * when called by nfs_readdir_add_to_array, the strings will be freed in
232 * nfs_clear_readdir_array()
235 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
238 string->name = kmemdup(name, len, GFP_KERNEL);
239 if (string->name == NULL)
242 * Avoid a kmemleak false positive. The pointer to the name is stored
243 * in a page cache page which kmemleak does not scan.
245 kmemleak_not_leak(string->name);
246 string->hash = full_name_hash(name, len);
251 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
253 struct nfs_cache_array *array = nfs_readdir_get_array(page);
254 struct nfs_cache_array_entry *cache_entry;
258 return PTR_ERR(array);
260 cache_entry = &array->array[array->size];
262 /* Check that this entry lies within the page bounds */
264 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
267 cache_entry->cookie = entry->prev_cookie;
268 cache_entry->ino = entry->ino;
269 cache_entry->d_type = entry->d_type;
270 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
273 array->last_cookie = entry->cookie;
276 array->eof_index = array->size;
278 nfs_readdir_release_array(page);
283 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
285 loff_t diff = desc->file->f_pos - desc->current_index;
290 if (diff >= array->size) {
291 if (array->eof_index >= 0)
293 desc->current_index += array->size;
297 index = (unsigned int)diff;
298 *desc->dir_cookie = array->array[index].cookie;
299 desc->cache_entry_index = index;
307 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
310 int status = -EAGAIN;
312 for (i = 0; i < array->size; i++) {
313 if (array->array[i].cookie == *desc->dir_cookie) {
314 desc->cache_entry_index = i;
319 if (i == array->eof_index) {
321 status = -EBADCOOKIE;
328 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
330 struct nfs_cache_array *array;
331 int status = -EBADCOOKIE;
333 if (desc->dir_cookie == NULL)
336 array = nfs_readdir_get_array(desc->page);
338 status = PTR_ERR(array);
342 if (*desc->dir_cookie == 0)
343 status = nfs_readdir_search_for_pos(array, desc);
345 status = nfs_readdir_search_for_cookie(array, desc);
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 rpc_cred *cred = nfs_file_cred(file);
358 unsigned long timestamp, gencount;
363 gencount = nfs_inc_attr_generation_counter();
364 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
365 NFS_SERVER(inode)->dtsize, desc->plus);
367 /* We requested READDIRPLUS, but the server doesn't grok it */
368 if (error == -ENOTSUPP && desc->plus) {
369 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
370 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
376 desc->timestamp = timestamp;
377 desc->gencount = gencount;
382 /* Fill in an entry based on the xdr code stored in desc->page */
384 int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream)
386 __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus);
390 entry->fattr->time_start = desc->timestamp;
391 entry->fattr->gencount = desc->gencount;
396 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
398 if (dentry->d_inode == NULL)
400 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
408 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
410 struct qstr filename = {
414 struct dentry *dentry;
415 struct dentry *alias;
416 struct inode *dir = parent->d_inode;
419 if (filename.name[0] == '.') {
420 if (filename.len == 1)
422 if (filename.len == 2 && filename.name[1] == '.')
425 filename.hash = full_name_hash(filename.name, filename.len);
427 dentry = d_lookup(parent, &filename);
428 if (dentry != NULL) {
429 if (nfs_same_file(dentry, entry)) {
430 nfs_refresh_inode(dentry->d_inode, entry->fattr);
438 dentry = d_alloc(parent, &filename);
442 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
443 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
447 alias = d_materialise_unique(dentry, inode);
451 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
454 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
460 /* Perform conversion from xdr to cache array */
462 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
463 void *xdr_page, struct page *page, unsigned int buflen)
465 struct xdr_stream stream;
467 __be32 *ptr = xdr_page;
468 struct nfs_cache_array *array;
469 unsigned int count = 0;
472 buf.head->iov_base = xdr_page;
473 buf.head->iov_len = buflen;
474 buf.tail->iov_len = 0;
477 buf.buflen = buf.head->iov_len;
478 buf.len = buf.head->iov_len;
480 xdr_init_decode(&stream, &buf, ptr);
484 status = xdr_decode(desc, entry, &stream);
486 if (status == -EAGAIN)
494 nfs_prime_dcache(desc->file->f_path.dentry, entry);
496 status = nfs_readdir_add_to_array(entry, page);
499 } while (!entry->eof);
501 if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) {
502 array = nfs_readdir_get_array(page);
503 if (!IS_ERR(array)) {
504 array->eof_index = array->size;
506 nfs_readdir_release_array(page);
508 status = PTR_ERR(array);
514 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
517 for (i = 0; i < npages; i++)
522 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
525 vm_unmap_ram(ptr, npages);
526 nfs_readdir_free_pagearray(pages, npages);
530 * nfs_readdir_large_page will allocate pages that must be freed with a call
531 * to nfs_readdir_free_large_page
534 void *nfs_readdir_large_page(struct page **pages, unsigned int npages)
539 for (i = 0; i < npages; i++) {
540 struct page *page = alloc_page(GFP_KERNEL);
546 ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL);
547 if (!IS_ERR_OR_NULL(ptr))
550 nfs_readdir_free_pagearray(pages, i);
555 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
557 struct page *pages[NFS_MAX_READDIR_PAGES];
558 void *pages_ptr = NULL;
559 struct nfs_entry entry;
560 struct file *file = desc->file;
561 struct nfs_cache_array *array;
562 int status = -ENOMEM;
563 unsigned int array_size = ARRAY_SIZE(pages);
565 entry.prev_cookie = 0;
566 entry.cookie = *desc->dir_cookie;
568 entry.fh = nfs_alloc_fhandle();
569 entry.fattr = nfs_alloc_fattr();
570 if (entry.fh == NULL || entry.fattr == NULL)
573 array = nfs_readdir_get_array(page);
575 status = PTR_ERR(array);
578 memset(array, 0, sizeof(struct nfs_cache_array));
579 array->eof_index = -1;
581 pages_ptr = nfs_readdir_large_page(pages, array_size);
583 goto out_release_array;
586 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
591 status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen);
593 if (status == -ENOSPC)
597 } while (array->eof_index < 0);
599 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
601 nfs_readdir_release_array(page);
603 nfs_free_fattr(entry.fattr);
604 nfs_free_fhandle(entry.fh);
609 * Now we cache directories properly, by converting xdr information
610 * to an array that can be used for lookups later. This results in
611 * fewer cache pages, since we can store more information on each page.
612 * We only need to convert from xdr once so future lookups are much simpler
615 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
617 struct inode *inode = desc->file->f_path.dentry->d_inode;
620 ret = nfs_readdir_xdr_to_array(desc, page, inode);
623 SetPageUptodate(page);
625 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
626 /* Should never happen */
627 nfs_zap_mapping(inode, inode->i_mapping);
637 void cache_page_release(nfs_readdir_descriptor_t *desc)
639 page_cache_release(desc->page);
644 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
646 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
647 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
651 * Returns 0 if desc->dir_cookie was found on page desc->page_index
654 int find_cache_page(nfs_readdir_descriptor_t *desc)
658 desc->page = get_cache_page(desc);
659 if (IS_ERR(desc->page))
660 return PTR_ERR(desc->page);
662 res = nfs_readdir_search_array(desc);
665 cache_page_release(desc);
669 /* Search for desc->dir_cookie from the beginning of the page cache */
671 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
675 if (desc->page_index == 0)
676 desc->current_index = 0;
678 res = find_cache_page(desc);
686 static inline unsigned int dt_type(struct inode *inode)
688 return (inode->i_mode >> 12) & 15;
692 * Once we've found the start of the dirent within a page: fill 'er up...
695 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
698 struct file *file = desc->file;
701 struct nfs_cache_array *array = NULL;
703 array = nfs_readdir_get_array(desc->page);
705 res = PTR_ERR(array);
709 for (i = desc->cache_entry_index; i < array->size; i++) {
710 struct nfs_cache_array_entry *ent;
712 ent = &array->array[i];
713 if (filldir(dirent, ent->string.name, ent->string.len,
714 file->f_pos, nfs_compat_user_ino64(ent->ino),
720 desc->cache_entry_index = i;
721 if (i < (array->size-1))
722 *desc->dir_cookie = array->array[i+1].cookie;
724 *desc->dir_cookie = array->last_cookie;
726 if (i == array->eof_index)
729 nfs_readdir_release_array(desc->page);
731 cache_page_release(desc);
732 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
733 (unsigned long long)*desc->dir_cookie, res);
738 * If we cannot find a cookie in our cache, we suspect that this is
739 * because it points to a deleted file, so we ask the server to return
740 * whatever it thinks is the next entry. We then feed this to filldir.
741 * If all goes well, we should then be able to find our way round the
742 * cache on the next call to readdir_search_pagecache();
744 * NOTE: we cannot add the anonymous page to the pagecache because
745 * the data it contains might not be page aligned. Besides,
746 * we should already have a complete representation of the
747 * directory in the page cache by the time we get here.
750 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
753 struct page *page = NULL;
755 struct inode *inode = desc->file->f_path.dentry->d_inode;
757 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
758 (unsigned long long)*desc->dir_cookie);
760 page = alloc_page(GFP_HIGHUSER);
766 desc->page_index = 0;
769 status = nfs_readdir_xdr_to_array(desc, page, inode);
773 status = nfs_do_filldir(desc, dirent, filldir);
776 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
780 cache_page_release(desc);
784 /* The file offset position represents the dirent entry number. A
785 last cookie cache takes care of the common case of reading the
788 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
790 struct dentry *dentry = filp->f_path.dentry;
791 struct inode *inode = dentry->d_inode;
792 nfs_readdir_descriptor_t my_desc,
796 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
797 dentry->d_parent->d_name.name, dentry->d_name.name,
798 (long long)filp->f_pos);
799 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
802 * filp->f_pos points to the dirent entry number.
803 * *desc->dir_cookie has the cookie for the next entry. We have
804 * to either find the entry with the appropriate number or
805 * revalidate the cookie.
807 memset(desc, 0, sizeof(*desc));
810 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
811 desc->decode = NFS_PROTO(inode)->decode_dirent;
812 desc->plus = NFS_USE_READDIRPLUS(inode);
814 nfs_block_sillyrename(dentry);
815 res = nfs_revalidate_mapping(inode, filp->f_mapping);
819 while (desc->eof != 1) {
820 res = readdir_search_pagecache(desc);
822 if (res == -EBADCOOKIE) {
824 /* This means either end of directory */
825 if (*desc->dir_cookie && desc->eof == 0) {
826 /* Or that the server has 'lost' a cookie */
827 res = uncached_readdir(desc, dirent, filldir);
833 if (res == -ETOOSMALL && desc->plus) {
834 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
835 nfs_zap_caches(inode);
836 desc->page_index = 0;
844 res = nfs_do_filldir(desc, dirent, filldir);
849 nfs_unblock_sillyrename(dentry);
852 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
853 dentry->d_parent->d_name.name, dentry->d_name.name,
858 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
860 struct dentry *dentry = filp->f_path.dentry;
861 struct inode *inode = dentry->d_inode;
863 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
864 dentry->d_parent->d_name.name,
868 mutex_lock(&inode->i_mutex);
871 offset += filp->f_pos;
879 if (offset != filp->f_pos) {
880 filp->f_pos = offset;
881 nfs_file_open_context(filp)->dir_cookie = 0;
884 mutex_unlock(&inode->i_mutex);
889 * All directory operations under NFS are synchronous, so fsync()
890 * is a dummy operation.
892 static int nfs_fsync_dir(struct file *filp, int datasync)
894 struct dentry *dentry = filp->f_path.dentry;
896 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
897 dentry->d_parent->d_name.name, dentry->d_name.name,
900 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
905 * nfs_force_lookup_revalidate - Mark the directory as having changed
906 * @dir - pointer to directory inode
908 * This forces the revalidation code in nfs_lookup_revalidate() to do a
909 * full lookup on all child dentries of 'dir' whenever a change occurs
910 * on the server that might have invalidated our dcache.
912 * The caller should be holding dir->i_lock
914 void nfs_force_lookup_revalidate(struct inode *dir)
916 NFS_I(dir)->cache_change_attribute++;
920 * A check for whether or not the parent directory has changed.
921 * In the case it has, we assume that the dentries are untrustworthy
922 * and may need to be looked up again.
924 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
928 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
930 if (!nfs_verify_change_attribute(dir, dentry->d_time))
932 /* Revalidate nfsi->cache_change_attribute before we declare a match */
933 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
935 if (!nfs_verify_change_attribute(dir, dentry->d_time))
941 * Return the intent data that applies to this particular path component
943 * Note that the current set of intents only apply to the very last
944 * component of the path.
945 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
947 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
949 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
951 return nd->flags & mask;
955 * Use intent information to check whether or not we're going to do
956 * an O_EXCL create using this path component.
958 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
960 if (NFS_PROTO(dir)->version == 2)
962 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
966 * Inode and filehandle revalidation for lookups.
968 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
969 * or if the intent information indicates that we're about to open this
970 * particular file and the "nocto" mount flag is not set.
974 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
976 struct nfs_server *server = NFS_SERVER(inode);
978 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
981 /* VFS wants an on-the-wire revalidation */
982 if (nd->flags & LOOKUP_REVAL)
984 /* This is an open(2) */
985 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
986 !(server->flags & NFS_MOUNT_NOCTO) &&
987 (S_ISREG(inode->i_mode) ||
988 S_ISDIR(inode->i_mode)))
992 return nfs_revalidate_inode(server, inode);
994 return __nfs_revalidate_inode(server, inode);
998 * We judge how long we want to trust negative
999 * dentries by looking at the parent inode mtime.
1001 * If parent mtime has changed, we revalidate, else we wait for a
1002 * period corresponding to the parent's attribute cache timeout value.
1005 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1006 struct nameidata *nd)
1008 /* Don't revalidate a negative dentry if we're creating a new file */
1009 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1011 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1013 return !nfs_check_verifier(dir, dentry);
1017 * This is called every time the dcache has a lookup hit,
1018 * and we should check whether we can really trust that
1021 * NOTE! The hit can be a negative hit too, don't assume
1024 * If the parent directory is seen to have changed, we throw out the
1025 * cached dentry and do a new lookup.
1027 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
1030 struct inode *inode;
1031 struct dentry *parent;
1032 struct nfs_fh *fhandle = NULL;
1033 struct nfs_fattr *fattr = NULL;
1036 parent = dget_parent(dentry);
1037 dir = parent->d_inode;
1038 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1039 inode = dentry->d_inode;
1042 if (nfs_neg_need_reval(dir, dentry, nd))
1047 if (is_bad_inode(inode)) {
1048 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1049 __func__, dentry->d_parent->d_name.name,
1050 dentry->d_name.name);
1054 if (nfs_have_delegation(inode, FMODE_READ))
1055 goto out_set_verifier;
1057 /* Force a full look up iff the parent directory has changed */
1058 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1059 if (nfs_lookup_verify_inode(inode, nd))
1060 goto out_zap_parent;
1064 if (NFS_STALE(inode))
1068 fhandle = nfs_alloc_fhandle();
1069 fattr = nfs_alloc_fattr();
1070 if (fhandle == NULL || fattr == NULL)
1073 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1076 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1078 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1081 nfs_free_fattr(fattr);
1082 nfs_free_fhandle(fhandle);
1084 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1087 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1088 __func__, dentry->d_parent->d_name.name,
1089 dentry->d_name.name);
1092 nfs_zap_caches(dir);
1094 nfs_mark_for_revalidate(dir);
1095 if (inode && S_ISDIR(inode->i_mode)) {
1096 /* Purge readdir caches. */
1097 nfs_zap_caches(inode);
1098 /* If we have submounts, don't unhash ! */
1099 if (have_submounts(dentry))
1101 if (dentry->d_flags & DCACHE_DISCONNECTED)
1103 shrink_dcache_parent(dentry);
1106 nfs_free_fattr(fattr);
1107 nfs_free_fhandle(fhandle);
1109 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1110 __func__, dentry->d_parent->d_name.name,
1111 dentry->d_name.name);
1114 nfs_free_fattr(fattr);
1115 nfs_free_fhandle(fhandle);
1117 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1118 __func__, dentry->d_parent->d_name.name,
1119 dentry->d_name.name, error);
1124 * This is called from dput() when d_count is going to 0.
1126 static int nfs_dentry_delete(struct dentry *dentry)
1128 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1129 dentry->d_parent->d_name.name, dentry->d_name.name,
1132 /* Unhash any dentry with a stale inode */
1133 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1136 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1137 /* Unhash it, so that ->d_iput() would be called */
1140 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1141 /* Unhash it, so that ancestors of killed async unlink
1142 * files will be cleaned up during umount */
1149 static void nfs_drop_nlink(struct inode *inode)
1151 spin_lock(&inode->i_lock);
1152 if (inode->i_nlink > 0)
1154 spin_unlock(&inode->i_lock);
1158 * Called when the dentry loses inode.
1159 * We use it to clean up silly-renamed files.
1161 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1163 if (S_ISDIR(inode->i_mode))
1164 /* drop any readdir cache as it could easily be old */
1165 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1167 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1169 nfs_complete_unlink(dentry, inode);
1174 const struct dentry_operations nfs_dentry_operations = {
1175 .d_revalidate = nfs_lookup_revalidate,
1176 .d_delete = nfs_dentry_delete,
1177 .d_iput = nfs_dentry_iput,
1180 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1183 struct dentry *parent;
1184 struct inode *inode = NULL;
1185 struct nfs_fh *fhandle = NULL;
1186 struct nfs_fattr *fattr = NULL;
1189 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1190 dentry->d_parent->d_name.name, dentry->d_name.name);
1191 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1193 res = ERR_PTR(-ENAMETOOLONG);
1194 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1197 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1200 * If we're doing an exclusive create, optimize away the lookup
1201 * but don't hash the dentry.
1203 if (nfs_is_exclusive_create(dir, nd)) {
1204 d_instantiate(dentry, NULL);
1209 res = ERR_PTR(-ENOMEM);
1210 fhandle = nfs_alloc_fhandle();
1211 fattr = nfs_alloc_fattr();
1212 if (fhandle == NULL || fattr == NULL)
1215 parent = dentry->d_parent;
1216 /* Protect against concurrent sillydeletes */
1217 nfs_block_sillyrename(parent);
1218 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1219 if (error == -ENOENT)
1222 res = ERR_PTR(error);
1223 goto out_unblock_sillyrename;
1225 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1226 res = (struct dentry *)inode;
1228 goto out_unblock_sillyrename;
1231 res = d_materialise_unique(dentry, inode);
1234 goto out_unblock_sillyrename;
1237 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1238 out_unblock_sillyrename:
1239 nfs_unblock_sillyrename(parent);
1241 nfs_free_fattr(fattr);
1242 nfs_free_fhandle(fhandle);
1246 #ifdef CONFIG_NFS_V4
1247 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1249 const struct dentry_operations nfs4_dentry_operations = {
1250 .d_revalidate = nfs_open_revalidate,
1251 .d_delete = nfs_dentry_delete,
1252 .d_iput = nfs_dentry_iput,
1256 * Use intent information to determine whether we need to substitute
1257 * the NFSv4-style stateful OPEN for the LOOKUP call
1259 static int is_atomic_open(struct nameidata *nd)
1261 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1263 /* NFS does not (yet) have a stateful open for directories */
1264 if (nd->flags & LOOKUP_DIRECTORY)
1266 /* Are we trying to write to a read only partition? */
1267 if (__mnt_is_readonly(nd->path.mnt) &&
1268 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1273 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1275 struct path path = {
1276 .mnt = nd->path.mnt,
1279 struct nfs_open_context *ctx;
1280 struct rpc_cred *cred;
1281 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1283 cred = rpc_lookup_cred();
1285 return ERR_CAST(cred);
1286 ctx = alloc_nfs_open_context(&path, cred, fmode);
1289 return ERR_PTR(-ENOMEM);
1293 static int do_open(struct inode *inode, struct file *filp)
1295 nfs_fscache_set_inode_cookie(inode, filp);
1299 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1304 /* If the open_intent is for execute, we have an extra check to make */
1305 if (ctx->mode & FMODE_EXEC) {
1306 ret = nfs_may_open(ctx->path.dentry->d_inode,
1308 nd->intent.open.flags);
1312 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1314 ret = PTR_ERR(filp);
1316 nfs_file_set_open_context(filp, ctx);
1318 put_nfs_open_context(ctx);
1322 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1324 struct nfs_open_context *ctx;
1326 struct dentry *res = NULL;
1327 struct inode *inode;
1331 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1332 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1334 /* Check that we are indeed trying to open this file */
1335 if (!is_atomic_open(nd))
1338 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1339 res = ERR_PTR(-ENAMETOOLONG);
1342 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1344 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1346 if (nd->flags & LOOKUP_EXCL) {
1347 d_instantiate(dentry, NULL);
1351 ctx = nameidata_to_nfs_open_context(dentry, nd);
1352 res = ERR_CAST(ctx);
1356 open_flags = nd->intent.open.flags;
1357 if (nd->flags & LOOKUP_CREATE) {
1358 attr.ia_mode = nd->intent.open.create_mode;
1359 attr.ia_valid = ATTR_MODE;
1360 if (!IS_POSIXACL(dir))
1361 attr.ia_mode &= ~current_umask();
1363 open_flags &= ~(O_EXCL | O_CREAT);
1367 /* Open the file on the server */
1368 nfs_block_sillyrename(dentry->d_parent);
1369 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1370 if (IS_ERR(inode)) {
1371 nfs_unblock_sillyrename(dentry->d_parent);
1372 put_nfs_open_context(ctx);
1373 switch (PTR_ERR(inode)) {
1374 /* Make a negative dentry */
1376 d_add(dentry, NULL);
1379 /* This turned out not to be a regular file */
1383 if (!(nd->intent.open.flags & O_NOFOLLOW))
1388 res = ERR_CAST(inode);
1392 res = d_add_unique(dentry, inode);
1393 nfs_unblock_sillyrename(dentry->d_parent);
1395 dput(ctx->path.dentry);
1396 ctx->path.dentry = dget(res);
1399 err = nfs_intent_set_file(nd, ctx);
1403 return ERR_PTR(err);
1406 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1409 return nfs_lookup(dir, dentry, nd);
1412 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1414 struct dentry *parent = NULL;
1415 struct inode *inode = dentry->d_inode;
1417 struct nfs_open_context *ctx;
1418 int openflags, ret = 0;
1420 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1423 parent = dget_parent(dentry);
1424 dir = parent->d_inode;
1426 /* We can't create new files in nfs_open_revalidate(), so we
1427 * optimize away revalidation of negative dentries.
1429 if (inode == NULL) {
1430 if (!nfs_neg_need_reval(dir, dentry, nd))
1435 /* NFS only supports OPEN on regular files */
1436 if (!S_ISREG(inode->i_mode))
1438 openflags = nd->intent.open.flags;
1439 /* We cannot do exclusive creation on a positive dentry */
1440 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1442 /* We can't create new files, or truncate existing ones here */
1443 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1445 ctx = nameidata_to_nfs_open_context(dentry, nd);
1450 * Note: we're not holding inode->i_mutex and so may be racing with
1451 * operations that change the directory. We therefore save the
1452 * change attribute *before* we do the RPC call.
1454 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1455 if (IS_ERR(inode)) {
1456 ret = PTR_ERR(inode);
1469 if (inode != dentry->d_inode)
1472 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1473 ret = nfs_intent_set_file(nd, ctx);
1483 put_nfs_open_context(ctx);
1489 return nfs_lookup_revalidate(dentry, nd);
1492 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1493 struct nameidata *nd)
1495 struct nfs_open_context *ctx = NULL;
1500 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1501 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1503 attr.ia_mode = mode;
1504 attr.ia_valid = ATTR_MODE;
1506 if ((nd->flags & LOOKUP_CREATE) != 0) {
1507 open_flags = nd->intent.open.flags;
1509 ctx = nameidata_to_nfs_open_context(dentry, nd);
1510 error = PTR_ERR(ctx);
1515 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1519 error = nfs_intent_set_file(nd, ctx);
1526 put_nfs_open_context(ctx);
1533 #endif /* CONFIG_NFSV4 */
1536 * Code common to create, mkdir, and mknod.
1538 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1539 struct nfs_fattr *fattr)
1541 struct dentry *parent = dget_parent(dentry);
1542 struct inode *dir = parent->d_inode;
1543 struct inode *inode;
1544 int error = -EACCES;
1548 /* We may have been initialized further down */
1549 if (dentry->d_inode)
1551 if (fhandle->size == 0) {
1552 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1556 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1557 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1558 struct nfs_server *server = NFS_SB(dentry->d_sb);
1559 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1563 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1564 error = PTR_ERR(inode);
1567 d_add(dentry, inode);
1572 nfs_mark_for_revalidate(dir);
1578 * Following a failed create operation, we drop the dentry rather
1579 * than retain a negative dentry. This avoids a problem in the event
1580 * that the operation succeeded on the server, but an error in the
1581 * reply path made it appear to have failed.
1583 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1584 struct nameidata *nd)
1589 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1590 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1592 attr.ia_mode = mode;
1593 attr.ia_valid = ATTR_MODE;
1595 error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL);
1605 * See comments for nfs_proc_create regarding failed operations.
1608 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1613 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1614 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1616 if (!new_valid_dev(rdev))
1619 attr.ia_mode = mode;
1620 attr.ia_valid = ATTR_MODE;
1622 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1632 * See comments for nfs_proc_create regarding failed operations.
1634 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1639 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1640 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1642 attr.ia_valid = ATTR_MODE;
1643 attr.ia_mode = mode | S_IFDIR;
1645 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1654 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1656 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1660 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1664 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1665 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1667 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1668 /* Ensure the VFS deletes this inode */
1669 if (error == 0 && dentry->d_inode != NULL)
1670 clear_nlink(dentry->d_inode);
1671 else if (error == -ENOENT)
1672 nfs_dentry_handle_enoent(dentry);
1678 * Remove a file after making sure there are no pending writes,
1679 * and after checking that the file has only one user.
1681 * We invalidate the attribute cache and free the inode prior to the operation
1682 * to avoid possible races if the server reuses the inode.
1684 static int nfs_safe_remove(struct dentry *dentry)
1686 struct inode *dir = dentry->d_parent->d_inode;
1687 struct inode *inode = dentry->d_inode;
1690 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1691 dentry->d_parent->d_name.name, dentry->d_name.name);
1693 /* If the dentry was sillyrenamed, we simply call d_delete() */
1694 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1699 if (inode != NULL) {
1700 nfs_inode_return_delegation(inode);
1701 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1702 /* The VFS may want to delete this inode */
1704 nfs_drop_nlink(inode);
1705 nfs_mark_for_revalidate(inode);
1707 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1708 if (error == -ENOENT)
1709 nfs_dentry_handle_enoent(dentry);
1714 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1715 * belongs to an active ".nfs..." file and we return -EBUSY.
1717 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1719 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1722 int need_rehash = 0;
1724 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1725 dir->i_ino, dentry->d_name.name);
1727 spin_lock(&dcache_lock);
1728 spin_lock(&dentry->d_lock);
1729 if (atomic_read(&dentry->d_count) > 1) {
1730 spin_unlock(&dentry->d_lock);
1731 spin_unlock(&dcache_lock);
1732 /* Start asynchronous writeout of the inode */
1733 write_inode_now(dentry->d_inode, 0);
1734 error = nfs_sillyrename(dir, dentry);
1737 if (!d_unhashed(dentry)) {
1741 spin_unlock(&dentry->d_lock);
1742 spin_unlock(&dcache_lock);
1743 error = nfs_safe_remove(dentry);
1744 if (!error || error == -ENOENT) {
1745 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1746 } else if (need_rehash)
1752 * To create a symbolic link, most file systems instantiate a new inode,
1753 * add a page to it containing the path, then write it out to the disk
1754 * using prepare_write/commit_write.
1756 * Unfortunately the NFS client can't create the in-core inode first
1757 * because it needs a file handle to create an in-core inode (see
1758 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1759 * symlink request has completed on the server.
1761 * So instead we allocate a raw page, copy the symname into it, then do
1762 * the SYMLINK request with the page as the buffer. If it succeeds, we
1763 * now have a new file handle and can instantiate an in-core NFS inode
1764 * and move the raw page into its mapping.
1766 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1768 struct pagevec lru_pvec;
1772 unsigned int pathlen = strlen(symname);
1775 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1776 dir->i_ino, dentry->d_name.name, symname);
1778 if (pathlen > PAGE_SIZE)
1779 return -ENAMETOOLONG;
1781 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1782 attr.ia_valid = ATTR_MODE;
1784 page = alloc_page(GFP_HIGHUSER);
1788 kaddr = kmap_atomic(page, KM_USER0);
1789 memcpy(kaddr, symname, pathlen);
1790 if (pathlen < PAGE_SIZE)
1791 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1792 kunmap_atomic(kaddr, KM_USER0);
1794 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1796 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1797 dir->i_sb->s_id, dir->i_ino,
1798 dentry->d_name.name, symname, error);
1805 * No big deal if we can't add this page to the page cache here.
1806 * READLINK will get the missing page from the server if needed.
1808 pagevec_init(&lru_pvec, 0);
1809 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1811 pagevec_add(&lru_pvec, page);
1812 pagevec_lru_add_file(&lru_pvec);
1813 SetPageUptodate(page);
1822 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1824 struct inode *inode = old_dentry->d_inode;
1827 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1828 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1829 dentry->d_parent->d_name.name, dentry->d_name.name);
1831 nfs_inode_return_delegation(inode);
1834 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1837 d_add(dentry, inode);
1844 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1845 * different file handle for the same inode after a rename (e.g. when
1846 * moving to a different directory). A fail-safe method to do so would
1847 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1848 * rename the old file using the sillyrename stuff. This way, the original
1849 * file in old_dir will go away when the last process iput()s the inode.
1853 * It actually works quite well. One needs to have the possibility for
1854 * at least one ".nfs..." file in each directory the file ever gets
1855 * moved or linked to which happens automagically with the new
1856 * implementation that only depends on the dcache stuff instead of
1857 * using the inode layer
1859 * Unfortunately, things are a little more complicated than indicated
1860 * above. For a cross-directory move, we want to make sure we can get
1861 * rid of the old inode after the operation. This means there must be
1862 * no pending writes (if it's a file), and the use count must be 1.
1863 * If these conditions are met, we can drop the dentries before doing
1866 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1867 struct inode *new_dir, struct dentry *new_dentry)
1869 struct inode *old_inode = old_dentry->d_inode;
1870 struct inode *new_inode = new_dentry->d_inode;
1871 struct dentry *dentry = NULL, *rehash = NULL;
1874 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1875 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1876 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1877 atomic_read(&new_dentry->d_count));
1880 * For non-directories, check whether the target is busy and if so,
1881 * make a copy of the dentry and then do a silly-rename. If the
1882 * silly-rename succeeds, the copied dentry is hashed and becomes
1885 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1887 * To prevent any new references to the target during the
1888 * rename, we unhash the dentry in advance.
1890 if (!d_unhashed(new_dentry)) {
1892 rehash = new_dentry;
1895 if (atomic_read(&new_dentry->d_count) > 2) {
1898 /* copy the target dentry's name */
1899 dentry = d_alloc(new_dentry->d_parent,
1900 &new_dentry->d_name);
1904 /* silly-rename the existing target ... */
1905 err = nfs_sillyrename(new_dir, new_dentry);
1909 new_dentry = dentry;
1915 nfs_inode_return_delegation(old_inode);
1916 if (new_inode != NULL)
1917 nfs_inode_return_delegation(new_inode);
1919 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1920 new_dir, &new_dentry->d_name);
1921 nfs_mark_for_revalidate(old_inode);
1926 if (new_inode != NULL)
1927 nfs_drop_nlink(new_inode);
1928 d_move(old_dentry, new_dentry);
1929 nfs_set_verifier(new_dentry,
1930 nfs_save_change_attribute(new_dir));
1931 } else if (error == -ENOENT)
1932 nfs_dentry_handle_enoent(old_dentry);
1934 /* new dentry created? */
1940 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1941 static LIST_HEAD(nfs_access_lru_list);
1942 static atomic_long_t nfs_access_nr_entries;
1944 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1946 put_rpccred(entry->cred);
1948 smp_mb__before_atomic_dec();
1949 atomic_long_dec(&nfs_access_nr_entries);
1950 smp_mb__after_atomic_dec();
1953 static void nfs_access_free_list(struct list_head *head)
1955 struct nfs_access_entry *cache;
1957 while (!list_empty(head)) {
1958 cache = list_entry(head->next, struct nfs_access_entry, lru);
1959 list_del(&cache->lru);
1960 nfs_access_free_entry(cache);
1964 int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask)
1967 struct nfs_inode *nfsi, *next;
1968 struct nfs_access_entry *cache;
1970 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1971 return (nr_to_scan == 0) ? 0 : -1;
1973 spin_lock(&nfs_access_lru_lock);
1974 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1975 struct inode *inode;
1977 if (nr_to_scan-- == 0)
1979 inode = &nfsi->vfs_inode;
1980 spin_lock(&inode->i_lock);
1981 if (list_empty(&nfsi->access_cache_entry_lru))
1982 goto remove_lru_entry;
1983 cache = list_entry(nfsi->access_cache_entry_lru.next,
1984 struct nfs_access_entry, lru);
1985 list_move(&cache->lru, &head);
1986 rb_erase(&cache->rb_node, &nfsi->access_cache);
1987 if (!list_empty(&nfsi->access_cache_entry_lru))
1988 list_move_tail(&nfsi->access_cache_inode_lru,
1989 &nfs_access_lru_list);
1992 list_del_init(&nfsi->access_cache_inode_lru);
1993 smp_mb__before_clear_bit();
1994 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1995 smp_mb__after_clear_bit();
1997 spin_unlock(&inode->i_lock);
1999 spin_unlock(&nfs_access_lru_lock);
2000 nfs_access_free_list(&head);
2001 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2004 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2006 struct rb_root *root_node = &nfsi->access_cache;
2008 struct nfs_access_entry *entry;
2010 /* Unhook entries from the cache */
2011 while ((n = rb_first(root_node)) != NULL) {
2012 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2013 rb_erase(n, root_node);
2014 list_move(&entry->lru, head);
2016 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2019 void nfs_access_zap_cache(struct inode *inode)
2023 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2025 /* Remove from global LRU init */
2026 spin_lock(&nfs_access_lru_lock);
2027 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2028 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2030 spin_lock(&inode->i_lock);
2031 __nfs_access_zap_cache(NFS_I(inode), &head);
2032 spin_unlock(&inode->i_lock);
2033 spin_unlock(&nfs_access_lru_lock);
2034 nfs_access_free_list(&head);
2037 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2039 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2040 struct nfs_access_entry *entry;
2043 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2045 if (cred < entry->cred)
2047 else if (cred > entry->cred)
2055 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2057 struct nfs_inode *nfsi = NFS_I(inode);
2058 struct nfs_access_entry *cache;
2061 spin_lock(&inode->i_lock);
2062 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2064 cache = nfs_access_search_rbtree(inode, cred);
2067 if (!nfs_have_delegated_attributes(inode) &&
2068 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2070 res->jiffies = cache->jiffies;
2071 res->cred = cache->cred;
2072 res->mask = cache->mask;
2073 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2076 spin_unlock(&inode->i_lock);
2079 rb_erase(&cache->rb_node, &nfsi->access_cache);
2080 list_del(&cache->lru);
2081 spin_unlock(&inode->i_lock);
2082 nfs_access_free_entry(cache);
2085 spin_unlock(&inode->i_lock);
2086 nfs_access_zap_cache(inode);
2090 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2092 struct nfs_inode *nfsi = NFS_I(inode);
2093 struct rb_root *root_node = &nfsi->access_cache;
2094 struct rb_node **p = &root_node->rb_node;
2095 struct rb_node *parent = NULL;
2096 struct nfs_access_entry *entry;
2098 spin_lock(&inode->i_lock);
2099 while (*p != NULL) {
2101 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2103 if (set->cred < entry->cred)
2104 p = &parent->rb_left;
2105 else if (set->cred > entry->cred)
2106 p = &parent->rb_right;
2110 rb_link_node(&set->rb_node, parent, p);
2111 rb_insert_color(&set->rb_node, root_node);
2112 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2113 spin_unlock(&inode->i_lock);
2116 rb_replace_node(parent, &set->rb_node, root_node);
2117 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2118 list_del(&entry->lru);
2119 spin_unlock(&inode->i_lock);
2120 nfs_access_free_entry(entry);
2123 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2125 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2128 RB_CLEAR_NODE(&cache->rb_node);
2129 cache->jiffies = set->jiffies;
2130 cache->cred = get_rpccred(set->cred);
2131 cache->mask = set->mask;
2133 nfs_access_add_rbtree(inode, cache);
2135 /* Update accounting */
2136 smp_mb__before_atomic_inc();
2137 atomic_long_inc(&nfs_access_nr_entries);
2138 smp_mb__after_atomic_inc();
2140 /* Add inode to global LRU list */
2141 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2142 spin_lock(&nfs_access_lru_lock);
2143 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2144 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2145 &nfs_access_lru_list);
2146 spin_unlock(&nfs_access_lru_lock);
2150 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2152 struct nfs_access_entry cache;
2155 status = nfs_access_get_cached(inode, cred, &cache);
2159 /* Be clever: ask server to check for all possible rights */
2160 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2162 cache.jiffies = jiffies;
2163 status = NFS_PROTO(inode)->access(inode, &cache);
2165 if (status == -ESTALE) {
2166 nfs_zap_caches(inode);
2167 if (!S_ISDIR(inode->i_mode))
2168 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2172 nfs_access_add_cache(inode, &cache);
2174 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2179 static int nfs_open_permission_mask(int openflags)
2183 if (openflags & FMODE_READ)
2185 if (openflags & FMODE_WRITE)
2187 if (openflags & FMODE_EXEC)
2192 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2194 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2197 int nfs_permission(struct inode *inode, int mask)
2199 struct rpc_cred *cred;
2202 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2204 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2206 /* Is this sys_access() ? */
2207 if (mask & (MAY_ACCESS | MAY_CHDIR))
2210 switch (inode->i_mode & S_IFMT) {
2214 /* NFSv4 has atomic_open... */
2215 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2216 && (mask & MAY_OPEN)
2217 && !(mask & MAY_EXEC))
2222 * Optimize away all write operations, since the server
2223 * will check permissions when we perform the op.
2225 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2230 if (!NFS_PROTO(inode)->access)
2233 cred = rpc_lookup_cred();
2234 if (!IS_ERR(cred)) {
2235 res = nfs_do_access(inode, cred, mask);
2238 res = PTR_ERR(cred);
2240 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2243 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2244 inode->i_sb->s_id, inode->i_ino, mask, res);
2247 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2249 res = generic_permission(inode, mask, NULL);
2255 * version-control: t
2256 * kept-new-versions: 5