4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/pagevec.h>
33 #include <linux/namei.h>
34 #include <linux/mount.h>
35 #include <linux/sched.h>
36 #include <linux/kmemleak.h>
37 #include <linux/xattr.h>
39 #include "delegation.h"
44 /* #define NFS_DEBUG_VERBOSE 1 */
46 static int nfs_opendir(struct inode *, struct file *);
47 static int nfs_closedir(struct inode *, struct file *);
48 static int nfs_readdir(struct file *, void *, filldir_t);
49 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, int);
52 static int nfs_rmdir(struct inode *, struct dentry *);
53 static int nfs_unlink(struct inode *, struct dentry *);
54 static int nfs_symlink(struct inode *, struct dentry *, const char *);
55 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, int);
60 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61 static void nfs_readdir_clear_array(struct page*);
63 const struct file_operations nfs_dir_operations = {
64 .llseek = nfs_llseek_dir,
65 .read = generic_read_dir,
66 .readdir = nfs_readdir,
68 .release = nfs_closedir,
69 .fsync = nfs_fsync_dir,
72 const struct inode_operations nfs_dir_inode_operations = {
77 .symlink = nfs_symlink,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
87 const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
92 const struct inode_operations nfs3_dir_inode_operations = {
97 .symlink = nfs_symlink,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
110 #endif /* CONFIG_NFS_V3 */
114 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
116 const struct inode_operations nfs4_dir_inode_operations = {
117 .create = nfs_open_create,
118 .lookup = nfs_atomic_lookup,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
135 #endif /* CONFIG_NFS_V4 */
137 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
139 struct nfs_open_dir_context *ctx;
140 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
143 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
146 ctx->cred = get_rpccred(cred);
149 return ERR_PTR(-ENOMEM);
152 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
154 put_rpccred(ctx->cred);
162 nfs_opendir(struct inode *inode, struct file *filp)
165 struct nfs_open_dir_context *ctx;
166 struct rpc_cred *cred;
168 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
169 filp->f_path.dentry->d_parent->d_name.name,
170 filp->f_path.dentry->d_name.name);
172 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
174 cred = rpc_lookup_cred();
176 return PTR_ERR(cred);
177 ctx = alloc_nfs_open_dir_context(inode, cred);
182 filp->private_data = ctx;
183 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
184 /* This is a mountpoint, so d_revalidate will never
185 * have been called, so we need to refresh the
186 * inode (for close-open consistency) ourselves.
188 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
196 nfs_closedir(struct inode *inode, struct file *filp)
198 put_nfs_open_dir_context(filp->private_data);
202 struct nfs_cache_array_entry {
206 unsigned char d_type;
209 struct nfs_cache_array {
213 struct nfs_cache_array_entry array[0];
216 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
220 unsigned long page_index;
223 loff_t current_index;
224 decode_dirent_t decode;
226 unsigned long timestamp;
227 unsigned long gencount;
228 unsigned int cache_entry_index;
231 } nfs_readdir_descriptor_t;
234 * The caller is responsible for calling nfs_readdir_release_array(page)
237 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
241 return ERR_PTR(-EIO);
244 return ERR_PTR(-ENOMEM);
249 void nfs_readdir_release_array(struct page *page)
255 * we are freeing strings created by nfs_add_to_readdir_array()
258 void nfs_readdir_clear_array(struct page *page)
260 struct nfs_cache_array *array;
263 array = kmap_atomic(page, KM_USER0);
264 for (i = 0; i < array->size; i++)
265 kfree(array->array[i].string.name);
266 kunmap_atomic(array, KM_USER0);
270 * the caller is responsible for freeing qstr.name
271 * when called by nfs_readdir_add_to_array, the strings will be freed in
272 * nfs_clear_readdir_array()
275 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
278 string->name = kmemdup(name, len, GFP_KERNEL);
279 if (string->name == NULL)
282 * Avoid a kmemleak false positive. The pointer to the name is stored
283 * in a page cache page which kmemleak does not scan.
285 kmemleak_not_leak(string->name);
286 string->hash = full_name_hash(name, len);
291 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
293 struct nfs_cache_array *array = nfs_readdir_get_array(page);
294 struct nfs_cache_array_entry *cache_entry;
298 return PTR_ERR(array);
300 cache_entry = &array->array[array->size];
302 /* Check that this entry lies within the page bounds */
304 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
307 cache_entry->cookie = entry->prev_cookie;
308 cache_entry->ino = entry->ino;
309 cache_entry->d_type = entry->d_type;
310 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
313 array->last_cookie = entry->cookie;
316 array->eof_index = array->size;
318 nfs_readdir_release_array(page);
323 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
325 loff_t diff = desc->file->f_pos - desc->current_index;
330 if (diff >= array->size) {
331 if (array->eof_index >= 0)
336 index = (unsigned int)diff;
337 *desc->dir_cookie = array->array[index].cookie;
338 desc->cache_entry_index = index;
346 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
350 int status = -EAGAIN;
352 for (i = 0; i < array->size; i++) {
353 if (array->array[i].cookie == *desc->dir_cookie) {
354 struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
355 struct nfs_open_dir_context *ctx = desc->file->private_data;
357 new_pos = desc->current_index + i;
358 if (ctx->attr_gencount != nfsi->attr_gencount
359 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
361 ctx->attr_gencount = nfsi->attr_gencount;
362 } else if (new_pos < desc->file->f_pos) {
364 && ctx->dup_cookie == *desc->dir_cookie) {
365 if (printk_ratelimit()) {
366 pr_notice("NFS: directory %s/%s contains a readdir loop."
367 "Please contact your server vendor. "
368 "Offending cookie: %llu\n",
369 desc->file->f_dentry->d_parent->d_name.name,
370 desc->file->f_dentry->d_name.name,
376 ctx->dup_cookie = *desc->dir_cookie;
379 desc->file->f_pos = new_pos;
380 desc->cache_entry_index = i;
384 if (array->eof_index >= 0) {
385 status = -EBADCOOKIE;
386 if (*desc->dir_cookie == array->last_cookie)
394 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
396 struct nfs_cache_array *array;
399 array = nfs_readdir_get_array(desc->page);
401 status = PTR_ERR(array);
405 if (*desc->dir_cookie == 0)
406 status = nfs_readdir_search_for_pos(array, desc);
408 status = nfs_readdir_search_for_cookie(array, desc);
410 if (status == -EAGAIN) {
411 desc->last_cookie = array->last_cookie;
412 desc->current_index += array->size;
415 nfs_readdir_release_array(desc->page);
420 /* Fill a page with xdr information before transferring to the cache page */
422 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
423 struct nfs_entry *entry, struct file *file, struct inode *inode)
425 struct nfs_open_dir_context *ctx = file->private_data;
426 struct rpc_cred *cred = ctx->cred;
427 unsigned long timestamp, gencount;
432 gencount = nfs_inc_attr_generation_counter();
433 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
434 NFS_SERVER(inode)->dtsize, desc->plus);
436 /* We requested READDIRPLUS, but the server doesn't grok it */
437 if (error == -ENOTSUPP && desc->plus) {
438 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
439 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
445 desc->timestamp = timestamp;
446 desc->gencount = gencount;
451 static int xdr_decode(nfs_readdir_descriptor_t *desc,
452 struct nfs_entry *entry, struct xdr_stream *xdr)
456 error = desc->decode(xdr, entry, desc->plus);
459 entry->fattr->time_start = desc->timestamp;
460 entry->fattr->gencount = desc->gencount;
465 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
467 if (dentry->d_inode == NULL)
469 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
477 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
479 struct qstr filename = {
483 struct dentry *dentry;
484 struct dentry *alias;
485 struct inode *dir = parent->d_inode;
488 if (filename.name[0] == '.') {
489 if (filename.len == 1)
491 if (filename.len == 2 && filename.name[1] == '.')
494 filename.hash = full_name_hash(filename.name, filename.len);
496 dentry = d_lookup(parent, &filename);
497 if (dentry != NULL) {
498 if (nfs_same_file(dentry, entry)) {
499 nfs_refresh_inode(dentry->d_inode, entry->fattr);
507 dentry = d_alloc(parent, &filename);
511 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
515 alias = d_materialise_unique(dentry, inode);
519 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
522 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
528 /* Perform conversion from xdr to cache array */
530 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
531 struct page **xdr_pages, struct page *page, unsigned int buflen)
533 struct xdr_stream stream;
535 struct page *scratch;
536 struct nfs_cache_array *array;
537 unsigned int count = 0;
540 scratch = alloc_page(GFP_KERNEL);
544 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
545 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
548 status = xdr_decode(desc, entry, &stream);
550 if (status == -EAGAIN)
558 nfs_prime_dcache(desc->file->f_path.dentry, entry);
560 status = nfs_readdir_add_to_array(entry, page);
563 } while (!entry->eof);
565 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
566 array = nfs_readdir_get_array(page);
567 if (!IS_ERR(array)) {
568 array->eof_index = array->size;
570 nfs_readdir_release_array(page);
572 status = PTR_ERR(array);
580 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
583 for (i = 0; i < npages; i++)
588 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
591 nfs_readdir_free_pagearray(pages, npages);
595 * nfs_readdir_large_page will allocate pages that must be freed with a call
596 * to nfs_readdir_free_large_page
599 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
603 for (i = 0; i < npages; i++) {
604 struct page *page = alloc_page(GFP_KERNEL);
612 nfs_readdir_free_pagearray(pages, i);
617 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
619 struct page *pages[NFS_MAX_READDIR_PAGES];
620 void *pages_ptr = NULL;
621 struct nfs_entry entry;
622 struct file *file = desc->file;
623 struct nfs_cache_array *array;
624 int status = -ENOMEM;
625 unsigned int array_size = ARRAY_SIZE(pages);
627 entry.prev_cookie = 0;
628 entry.cookie = desc->last_cookie;
630 entry.fh = nfs_alloc_fhandle();
631 entry.fattr = nfs_alloc_fattr();
632 entry.server = NFS_SERVER(inode);
633 if (entry.fh == NULL || entry.fattr == NULL)
636 array = nfs_readdir_get_array(page);
638 status = PTR_ERR(array);
641 memset(array, 0, sizeof(struct nfs_cache_array));
642 array->eof_index = -1;
644 status = nfs_readdir_large_page(pages, array_size);
646 goto out_release_array;
649 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
654 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
656 if (status == -ENOSPC)
660 } while (array->eof_index < 0);
662 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
664 nfs_readdir_release_array(page);
666 nfs_free_fattr(entry.fattr);
667 nfs_free_fhandle(entry.fh);
672 * Now we cache directories properly, by converting xdr information
673 * to an array that can be used for lookups later. This results in
674 * fewer cache pages, since we can store more information on each page.
675 * We only need to convert from xdr once so future lookups are much simpler
678 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
680 struct inode *inode = desc->file->f_path.dentry->d_inode;
683 ret = nfs_readdir_xdr_to_array(desc, page, inode);
686 SetPageUptodate(page);
688 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
689 /* Should never happen */
690 nfs_zap_mapping(inode, inode->i_mapping);
700 void cache_page_release(nfs_readdir_descriptor_t *desc)
702 if (!desc->page->mapping)
703 nfs_readdir_clear_array(desc->page);
704 page_cache_release(desc->page);
709 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
711 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
712 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
716 * Returns 0 if desc->dir_cookie was found on page desc->page_index
719 int find_cache_page(nfs_readdir_descriptor_t *desc)
723 desc->page = get_cache_page(desc);
724 if (IS_ERR(desc->page))
725 return PTR_ERR(desc->page);
727 res = nfs_readdir_search_array(desc);
729 cache_page_release(desc);
733 /* Search for desc->dir_cookie from the beginning of the page cache */
735 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
739 if (desc->page_index == 0) {
740 desc->current_index = 0;
741 desc->last_cookie = 0;
744 res = find_cache_page(desc);
745 } while (res == -EAGAIN);
750 * Once we've found the start of the dirent within a page: fill 'er up...
753 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
756 struct file *file = desc->file;
759 struct nfs_cache_array *array = NULL;
760 struct nfs_open_dir_context *ctx = file->private_data;
762 array = nfs_readdir_get_array(desc->page);
764 res = PTR_ERR(array);
768 for (i = desc->cache_entry_index; i < array->size; i++) {
769 struct nfs_cache_array_entry *ent;
771 ent = &array->array[i];
772 if (filldir(dirent, ent->string.name, ent->string.len,
773 file->f_pos, nfs_compat_user_ino64(ent->ino),
779 if (i < (array->size-1))
780 *desc->dir_cookie = array->array[i+1].cookie;
782 *desc->dir_cookie = array->last_cookie;
786 if (array->eof_index >= 0)
789 nfs_readdir_release_array(desc->page);
791 cache_page_release(desc);
792 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
793 (unsigned long long)*desc->dir_cookie, res);
798 * If we cannot find a cookie in our cache, we suspect that this is
799 * because it points to a deleted file, so we ask the server to return
800 * whatever it thinks is the next entry. We then feed this to filldir.
801 * If all goes well, we should then be able to find our way round the
802 * cache on the next call to readdir_search_pagecache();
804 * NOTE: we cannot add the anonymous page to the pagecache because
805 * the data it contains might not be page aligned. Besides,
806 * we should already have a complete representation of the
807 * directory in the page cache by the time we get here.
810 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
813 struct page *page = NULL;
815 struct inode *inode = desc->file->f_path.dentry->d_inode;
816 struct nfs_open_dir_context *ctx = desc->file->private_data;
818 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
819 (unsigned long long)*desc->dir_cookie);
821 page = alloc_page(GFP_HIGHUSER);
827 desc->page_index = 0;
828 desc->last_cookie = *desc->dir_cookie;
832 status = nfs_readdir_xdr_to_array(desc, page, inode);
836 status = nfs_do_filldir(desc, dirent, filldir);
839 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
843 cache_page_release(desc);
847 /* The file offset position represents the dirent entry number. A
848 last cookie cache takes care of the common case of reading the
851 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
853 struct dentry *dentry = filp->f_path.dentry;
854 struct inode *inode = dentry->d_inode;
855 nfs_readdir_descriptor_t my_desc,
857 struct nfs_open_dir_context *dir_ctx = filp->private_data;
860 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
861 dentry->d_parent->d_name.name, dentry->d_name.name,
862 (long long)filp->f_pos);
863 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
866 * filp->f_pos points to the dirent entry number.
867 * *desc->dir_cookie has the cookie for the next entry. We have
868 * to either find the entry with the appropriate number or
869 * revalidate the cookie.
871 memset(desc, 0, sizeof(*desc));
874 desc->dir_cookie = &dir_ctx->dir_cookie;
875 desc->decode = NFS_PROTO(inode)->decode_dirent;
876 desc->plus = NFS_USE_READDIRPLUS(inode);
878 nfs_block_sillyrename(dentry);
879 res = nfs_revalidate_mapping(inode, filp->f_mapping);
884 res = readdir_search_pagecache(desc);
886 if (res == -EBADCOOKIE) {
888 /* This means either end of directory */
889 if (*desc->dir_cookie && desc->eof == 0) {
890 /* Or that the server has 'lost' a cookie */
891 res = uncached_readdir(desc, dirent, filldir);
897 if (res == -ETOOSMALL && desc->plus) {
898 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
899 nfs_zap_caches(inode);
900 desc->page_index = 0;
908 res = nfs_do_filldir(desc, dirent, filldir);
911 } while (!desc->eof);
913 nfs_unblock_sillyrename(dentry);
916 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
917 dentry->d_parent->d_name.name, dentry->d_name.name,
922 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
924 struct dentry *dentry = filp->f_path.dentry;
925 struct inode *inode = dentry->d_inode;
926 struct nfs_open_dir_context *dir_ctx = filp->private_data;
928 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
929 dentry->d_parent->d_name.name,
933 mutex_lock(&inode->i_mutex);
936 offset += filp->f_pos;
944 if (offset != filp->f_pos) {
945 filp->f_pos = offset;
946 dir_ctx->dir_cookie = 0;
950 mutex_unlock(&inode->i_mutex);
955 * All directory operations under NFS are synchronous, so fsync()
956 * is a dummy operation.
958 static int nfs_fsync_dir(struct file *filp, int datasync)
960 struct dentry *dentry = filp->f_path.dentry;
962 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
963 dentry->d_parent->d_name.name, dentry->d_name.name,
966 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
971 * nfs_force_lookup_revalidate - Mark the directory as having changed
972 * @dir - pointer to directory inode
974 * This forces the revalidation code in nfs_lookup_revalidate() to do a
975 * full lookup on all child dentries of 'dir' whenever a change occurs
976 * on the server that might have invalidated our dcache.
978 * The caller should be holding dir->i_lock
980 void nfs_force_lookup_revalidate(struct inode *dir)
982 NFS_I(dir)->cache_change_attribute++;
986 * A check for whether or not the parent directory has changed.
987 * In the case it has, we assume that the dentries are untrustworthy
988 * and may need to be looked up again.
990 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
994 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
996 if (!nfs_verify_change_attribute(dir, dentry->d_time))
998 /* Revalidate nfsi->cache_change_attribute before we declare a match */
999 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1001 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1007 * Return the intent data that applies to this particular path component
1009 * Note that the current set of intents only apply to the very last
1010 * component of the path.
1011 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1013 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1016 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
1018 return nd->flags & mask;
1022 * Use intent information to check whether or not we're going to do
1023 * an O_EXCL create using this path component.
1025 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1027 if (NFS_PROTO(dir)->version == 2)
1029 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1033 * Inode and filehandle revalidation for lookups.
1035 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1036 * or if the intent information indicates that we're about to open this
1037 * particular file and the "nocto" mount flag is not set.
1041 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1043 struct nfs_server *server = NFS_SERVER(inode);
1045 if (IS_AUTOMOUNT(inode))
1048 /* VFS wants an on-the-wire revalidation */
1049 if (nd->flags & LOOKUP_REVAL)
1051 /* This is an open(2) */
1052 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1053 !(server->flags & NFS_MOUNT_NOCTO) &&
1054 (S_ISREG(inode->i_mode) ||
1055 S_ISDIR(inode->i_mode)))
1059 return nfs_revalidate_inode(server, inode);
1061 return __nfs_revalidate_inode(server, inode);
1065 * We judge how long we want to trust negative
1066 * dentries by looking at the parent inode mtime.
1068 * If parent mtime has changed, we revalidate, else we wait for a
1069 * period corresponding to the parent's attribute cache timeout value.
1072 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1073 struct nameidata *nd)
1075 /* Don't revalidate a negative dentry if we're creating a new file */
1076 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1078 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1080 return !nfs_check_verifier(dir, dentry);
1084 * This is called every time the dcache has a lookup hit,
1085 * and we should check whether we can really trust that
1088 * NOTE! The hit can be a negative hit too, don't assume
1091 * If the parent directory is seen to have changed, we throw out the
1092 * cached dentry and do a new lookup.
1094 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1097 struct inode *inode;
1098 struct dentry *parent;
1099 struct nfs_fh *fhandle = NULL;
1100 struct nfs_fattr *fattr = NULL;
1103 if (nd->flags & LOOKUP_RCU)
1106 parent = dget_parent(dentry);
1107 dir = parent->d_inode;
1108 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1109 inode = dentry->d_inode;
1112 if (nfs_neg_need_reval(dir, dentry, nd))
1117 if (is_bad_inode(inode)) {
1118 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1119 __func__, dentry->d_parent->d_name.name,
1120 dentry->d_name.name);
1124 if (nfs_have_delegation(inode, FMODE_READ))
1125 goto out_set_verifier;
1127 /* Force a full look up iff the parent directory has changed */
1128 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1129 if (nfs_lookup_verify_inode(inode, nd))
1130 goto out_zap_parent;
1134 if (NFS_STALE(inode))
1138 fhandle = nfs_alloc_fhandle();
1139 fattr = nfs_alloc_fattr();
1140 if (fhandle == NULL || fattr == NULL)
1143 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1146 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1148 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1151 nfs_free_fattr(fattr);
1152 nfs_free_fhandle(fhandle);
1154 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1157 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1158 __func__, dentry->d_parent->d_name.name,
1159 dentry->d_name.name);
1162 nfs_zap_caches(dir);
1164 nfs_mark_for_revalidate(dir);
1165 if (inode && S_ISDIR(inode->i_mode)) {
1166 /* Purge readdir caches. */
1167 nfs_zap_caches(inode);
1168 /* If we have submounts, don't unhash ! */
1169 if (have_submounts(dentry))
1171 if (dentry->d_flags & DCACHE_DISCONNECTED)
1173 shrink_dcache_parent(dentry);
1176 nfs_free_fattr(fattr);
1177 nfs_free_fhandle(fhandle);
1179 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1180 __func__, dentry->d_parent->d_name.name,
1181 dentry->d_name.name);
1184 nfs_free_fattr(fattr);
1185 nfs_free_fhandle(fhandle);
1187 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1188 __func__, dentry->d_parent->d_name.name,
1189 dentry->d_name.name, error);
1194 * This is called from dput() when d_count is going to 0.
1196 static int nfs_dentry_delete(const struct dentry *dentry)
1198 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1199 dentry->d_parent->d_name.name, dentry->d_name.name,
1202 /* Unhash any dentry with a stale inode */
1203 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1206 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1207 /* Unhash it, so that ->d_iput() would be called */
1210 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1211 /* Unhash it, so that ancestors of killed async unlink
1212 * files will be cleaned up during umount */
1219 static void nfs_drop_nlink(struct inode *inode)
1221 spin_lock(&inode->i_lock);
1222 if (inode->i_nlink > 0)
1224 spin_unlock(&inode->i_lock);
1228 * Called when the dentry loses inode.
1229 * We use it to clean up silly-renamed files.
1231 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1233 if (S_ISDIR(inode->i_mode))
1234 /* drop any readdir cache as it could easily be old */
1235 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1237 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1239 nfs_complete_unlink(dentry, inode);
1244 static void nfs_d_release(struct dentry *dentry)
1246 /* free cached devname value, if it survived that far */
1247 if (unlikely(dentry->d_fsdata)) {
1248 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1251 kfree(dentry->d_fsdata);
1255 const struct dentry_operations nfs_dentry_operations = {
1256 .d_revalidate = nfs_lookup_revalidate,
1257 .d_delete = nfs_dentry_delete,
1258 .d_iput = nfs_dentry_iput,
1259 .d_automount = nfs_d_automount,
1260 .d_release = nfs_d_release,
1263 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1266 struct dentry *parent;
1267 struct inode *inode = NULL;
1268 struct nfs_fh *fhandle = NULL;
1269 struct nfs_fattr *fattr = NULL;
1272 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1273 dentry->d_parent->d_name.name, dentry->d_name.name);
1274 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1276 res = ERR_PTR(-ENAMETOOLONG);
1277 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1281 * If we're doing an exclusive create, optimize away the lookup
1282 * but don't hash the dentry.
1284 if (nfs_is_exclusive_create(dir, nd)) {
1285 d_instantiate(dentry, NULL);
1290 res = ERR_PTR(-ENOMEM);
1291 fhandle = nfs_alloc_fhandle();
1292 fattr = nfs_alloc_fattr();
1293 if (fhandle == NULL || fattr == NULL)
1296 parent = dentry->d_parent;
1297 /* Protect against concurrent sillydeletes */
1298 nfs_block_sillyrename(parent);
1299 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1300 if (error == -ENOENT)
1303 res = ERR_PTR(error);
1304 goto out_unblock_sillyrename;
1306 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1307 res = ERR_CAST(inode);
1309 goto out_unblock_sillyrename;
1312 res = d_materialise_unique(dentry, inode);
1315 goto out_unblock_sillyrename;
1318 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1319 out_unblock_sillyrename:
1320 nfs_unblock_sillyrename(parent);
1322 nfs_free_fattr(fattr);
1323 nfs_free_fhandle(fhandle);
1327 #ifdef CONFIG_NFS_V4
1328 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1330 const struct dentry_operations nfs4_dentry_operations = {
1331 .d_revalidate = nfs_open_revalidate,
1332 .d_delete = nfs_dentry_delete,
1333 .d_iput = nfs_dentry_iput,
1334 .d_automount = nfs_d_automount,
1335 .d_release = nfs_d_release,
1339 * Use intent information to determine whether we need to substitute
1340 * the NFSv4-style stateful OPEN for the LOOKUP call
1342 static int is_atomic_open(struct nameidata *nd)
1344 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1346 /* NFS does not (yet) have a stateful open for directories */
1347 if (nd->flags & LOOKUP_DIRECTORY)
1349 /* Are we trying to write to a read only partition? */
1350 if (__mnt_is_readonly(nd->path.mnt) &&
1351 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1356 static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1358 struct path path = {
1359 .mnt = nd->path.mnt,
1362 struct nfs_open_context *ctx;
1363 struct rpc_cred *cred;
1364 fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1366 cred = rpc_lookup_cred();
1368 return ERR_CAST(cred);
1369 ctx = alloc_nfs_open_context(&path, cred, fmode);
1372 return ERR_PTR(-ENOMEM);
1376 static int do_open(struct inode *inode, struct file *filp)
1378 nfs_fscache_set_inode_cookie(inode, filp);
1382 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1387 /* If the open_intent is for execute, we have an extra check to make */
1388 if (ctx->mode & FMODE_EXEC) {
1389 ret = nfs_may_open(ctx->path.dentry->d_inode,
1391 nd->intent.open.flags);
1395 filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1397 ret = PTR_ERR(filp);
1399 nfs_file_set_open_context(filp, ctx);
1401 put_nfs_open_context(ctx);
1405 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1407 struct nfs_open_context *ctx;
1409 struct dentry *res = NULL;
1410 struct inode *inode;
1414 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1415 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1417 /* Check that we are indeed trying to open this file */
1418 if (!is_atomic_open(nd))
1421 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1422 res = ERR_PTR(-ENAMETOOLONG);
1426 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1428 if (nd->flags & LOOKUP_EXCL) {
1429 d_instantiate(dentry, NULL);
1433 ctx = nameidata_to_nfs_open_context(dentry, nd);
1434 res = ERR_CAST(ctx);
1438 open_flags = nd->intent.open.flags;
1439 if (nd->flags & LOOKUP_CREATE) {
1440 attr.ia_mode = nd->intent.open.create_mode;
1441 attr.ia_valid = ATTR_MODE;
1442 attr.ia_mode &= ~current_umask();
1444 open_flags &= ~(O_EXCL | O_CREAT);
1448 /* Open the file on the server */
1449 nfs_block_sillyrename(dentry->d_parent);
1450 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1451 if (IS_ERR(inode)) {
1452 nfs_unblock_sillyrename(dentry->d_parent);
1453 put_nfs_open_context(ctx);
1454 switch (PTR_ERR(inode)) {
1455 /* Make a negative dentry */
1457 d_add(dentry, NULL);
1460 /* This turned out not to be a regular file */
1465 if (!(nd->intent.open.flags & O_NOFOLLOW))
1469 res = ERR_CAST(inode);
1473 res = d_add_unique(dentry, inode);
1474 nfs_unblock_sillyrename(dentry->d_parent);
1476 dput(ctx->path.dentry);
1477 ctx->path.dentry = dget(res);
1480 err = nfs_intent_set_file(nd, ctx);
1484 return ERR_PTR(err);
1487 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1490 return nfs_lookup(dir, dentry, nd);
1493 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1495 struct dentry *parent = NULL;
1496 struct inode *inode;
1498 struct nfs_open_context *ctx;
1499 int openflags, ret = 0;
1501 if (nd->flags & LOOKUP_RCU)
1504 inode = dentry->d_inode;
1505 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1508 parent = dget_parent(dentry);
1509 dir = parent->d_inode;
1511 /* We can't create new files in nfs_open_revalidate(), so we
1512 * optimize away revalidation of negative dentries.
1514 if (inode == NULL) {
1515 if (!nfs_neg_need_reval(dir, dentry, nd))
1520 /* NFS only supports OPEN on regular files */
1521 if (!S_ISREG(inode->i_mode))
1523 openflags = nd->intent.open.flags;
1524 /* We cannot do exclusive creation on a positive dentry */
1525 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1527 /* We can't create new files, or truncate existing ones here */
1528 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1530 ctx = nameidata_to_nfs_open_context(dentry, nd);
1535 * Note: we're not holding inode->i_mutex and so may be racing with
1536 * operations that change the directory. We therefore save the
1537 * change attribute *before* we do the RPC call.
1539 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1540 if (IS_ERR(inode)) {
1541 ret = PTR_ERR(inode);
1554 if (inode != dentry->d_inode)
1557 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1558 ret = nfs_intent_set_file(nd, ctx);
1568 put_nfs_open_context(ctx);
1574 return nfs_lookup_revalidate(dentry, nd);
1577 static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1578 struct nameidata *nd)
1580 struct nfs_open_context *ctx = NULL;
1585 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1586 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1588 attr.ia_mode = mode;
1589 attr.ia_valid = ATTR_MODE;
1591 if ((nd->flags & LOOKUP_CREATE) != 0) {
1592 open_flags = nd->intent.open.flags;
1594 ctx = nameidata_to_nfs_open_context(dentry, nd);
1595 error = PTR_ERR(ctx);
1600 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1604 error = nfs_intent_set_file(nd, ctx);
1611 put_nfs_open_context(ctx);
1618 #endif /* CONFIG_NFSV4 */
1621 * Code common to create, mkdir, and mknod.
1623 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1624 struct nfs_fattr *fattr)
1626 struct dentry *parent = dget_parent(dentry);
1627 struct inode *dir = parent->d_inode;
1628 struct inode *inode;
1629 int error = -EACCES;
1633 /* We may have been initialized further down */
1634 if (dentry->d_inode)
1636 if (fhandle->size == 0) {
1637 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1641 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1642 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1643 struct nfs_server *server = NFS_SB(dentry->d_sb);
1644 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1648 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1649 error = PTR_ERR(inode);
1652 d_add(dentry, inode);
1657 nfs_mark_for_revalidate(dir);
1663 * Following a failed create operation, we drop the dentry rather
1664 * than retain a negative dentry. This avoids a problem in the event
1665 * that the operation succeeded on the server, but an error in the
1666 * reply path made it appear to have failed.
1668 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1669 struct nameidata *nd)
1675 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1676 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1678 attr.ia_mode = mode;
1679 attr.ia_valid = ATTR_MODE;
1681 if ((nd->flags & LOOKUP_CREATE) != 0)
1682 open_flags = nd->intent.open.flags;
1684 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1694 * See comments for nfs_proc_create regarding failed operations.
1697 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1702 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1703 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1705 if (!new_valid_dev(rdev))
1708 attr.ia_mode = mode;
1709 attr.ia_valid = ATTR_MODE;
1711 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1721 * See comments for nfs_proc_create regarding failed operations.
1723 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1728 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1729 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1731 attr.ia_valid = ATTR_MODE;
1732 attr.ia_mode = mode | S_IFDIR;
1734 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1743 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1745 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1749 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1753 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1754 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1756 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1757 /* Ensure the VFS deletes this inode */
1758 if (error == 0 && dentry->d_inode != NULL)
1759 clear_nlink(dentry->d_inode);
1760 else if (error == -ENOENT)
1761 nfs_dentry_handle_enoent(dentry);
1767 * Remove a file after making sure there are no pending writes,
1768 * and after checking that the file has only one user.
1770 * We invalidate the attribute cache and free the inode prior to the operation
1771 * to avoid possible races if the server reuses the inode.
1773 static int nfs_safe_remove(struct dentry *dentry)
1775 struct inode *dir = dentry->d_parent->d_inode;
1776 struct inode *inode = dentry->d_inode;
1779 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1780 dentry->d_parent->d_name.name, dentry->d_name.name);
1782 /* If the dentry was sillyrenamed, we simply call d_delete() */
1783 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1788 if (inode != NULL) {
1789 nfs_inode_return_delegation(inode);
1790 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1791 /* The VFS may want to delete this inode */
1793 nfs_drop_nlink(inode);
1794 nfs_mark_for_revalidate(inode);
1796 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1797 if (error == -ENOENT)
1798 nfs_dentry_handle_enoent(dentry);
1803 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1804 * belongs to an active ".nfs..." file and we return -EBUSY.
1806 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1808 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1811 int need_rehash = 0;
1813 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1814 dir->i_ino, dentry->d_name.name);
1816 spin_lock(&dentry->d_lock);
1817 if (dentry->d_count > 1) {
1818 spin_unlock(&dentry->d_lock);
1819 /* Start asynchronous writeout of the inode */
1820 write_inode_now(dentry->d_inode, 0);
1821 error = nfs_sillyrename(dir, dentry);
1824 if (!d_unhashed(dentry)) {
1828 spin_unlock(&dentry->d_lock);
1829 error = nfs_safe_remove(dentry);
1830 if (!error || error == -ENOENT) {
1831 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1832 } else if (need_rehash)
1838 * To create a symbolic link, most file systems instantiate a new inode,
1839 * add a page to it containing the path, then write it out to the disk
1840 * using prepare_write/commit_write.
1842 * Unfortunately the NFS client can't create the in-core inode first
1843 * because it needs a file handle to create an in-core inode (see
1844 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1845 * symlink request has completed on the server.
1847 * So instead we allocate a raw page, copy the symname into it, then do
1848 * the SYMLINK request with the page as the buffer. If it succeeds, we
1849 * now have a new file handle and can instantiate an in-core NFS inode
1850 * and move the raw page into its mapping.
1852 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1854 struct pagevec lru_pvec;
1858 unsigned int pathlen = strlen(symname);
1861 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1862 dir->i_ino, dentry->d_name.name, symname);
1864 if (pathlen > PAGE_SIZE)
1865 return -ENAMETOOLONG;
1867 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1868 attr.ia_valid = ATTR_MODE;
1870 page = alloc_page(GFP_HIGHUSER);
1874 kaddr = kmap_atomic(page, KM_USER0);
1875 memcpy(kaddr, symname, pathlen);
1876 if (pathlen < PAGE_SIZE)
1877 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1878 kunmap_atomic(kaddr, KM_USER0);
1880 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1882 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1883 dir->i_sb->s_id, dir->i_ino,
1884 dentry->d_name.name, symname, error);
1891 * No big deal if we can't add this page to the page cache here.
1892 * READLINK will get the missing page from the server if needed.
1894 pagevec_init(&lru_pvec, 0);
1895 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1897 pagevec_add(&lru_pvec, page);
1898 pagevec_lru_add_file(&lru_pvec);
1899 SetPageUptodate(page);
1908 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1910 struct inode *inode = old_dentry->d_inode;
1913 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1914 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1915 dentry->d_parent->d_name.name, dentry->d_name.name);
1917 nfs_inode_return_delegation(inode);
1920 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1923 d_add(dentry, inode);
1930 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1931 * different file handle for the same inode after a rename (e.g. when
1932 * moving to a different directory). A fail-safe method to do so would
1933 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1934 * rename the old file using the sillyrename stuff. This way, the original
1935 * file in old_dir will go away when the last process iput()s the inode.
1939 * It actually works quite well. One needs to have the possibility for
1940 * at least one ".nfs..." file in each directory the file ever gets
1941 * moved or linked to which happens automagically with the new
1942 * implementation that only depends on the dcache stuff instead of
1943 * using the inode layer
1945 * Unfortunately, things are a little more complicated than indicated
1946 * above. For a cross-directory move, we want to make sure we can get
1947 * rid of the old inode after the operation. This means there must be
1948 * no pending writes (if it's a file), and the use count must be 1.
1949 * If these conditions are met, we can drop the dentries before doing
1952 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1953 struct inode *new_dir, struct dentry *new_dentry)
1955 struct inode *old_inode = old_dentry->d_inode;
1956 struct inode *new_inode = new_dentry->d_inode;
1957 struct dentry *dentry = NULL, *rehash = NULL;
1960 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1961 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1962 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1963 new_dentry->d_count);
1966 * For non-directories, check whether the target is busy and if so,
1967 * make a copy of the dentry and then do a silly-rename. If the
1968 * silly-rename succeeds, the copied dentry is hashed and becomes
1971 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1973 * To prevent any new references to the target during the
1974 * rename, we unhash the dentry in advance.
1976 if (!d_unhashed(new_dentry)) {
1978 rehash = new_dentry;
1981 if (new_dentry->d_count > 2) {
1984 /* copy the target dentry's name */
1985 dentry = d_alloc(new_dentry->d_parent,
1986 &new_dentry->d_name);
1990 /* silly-rename the existing target ... */
1991 err = nfs_sillyrename(new_dir, new_dentry);
1995 new_dentry = dentry;
2001 nfs_inode_return_delegation(old_inode);
2002 if (new_inode != NULL)
2003 nfs_inode_return_delegation(new_inode);
2005 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2006 new_dir, &new_dentry->d_name);
2007 nfs_mark_for_revalidate(old_inode);
2012 if (new_inode != NULL)
2013 nfs_drop_nlink(new_inode);
2014 d_move(old_dentry, new_dentry);
2015 nfs_set_verifier(new_dentry,
2016 nfs_save_change_attribute(new_dir));
2017 } else if (error == -ENOENT)
2018 nfs_dentry_handle_enoent(old_dentry);
2020 /* new dentry created? */
2026 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2027 static LIST_HEAD(nfs_access_lru_list);
2028 static atomic_long_t nfs_access_nr_entries;
2030 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2032 put_rpccred(entry->cred);
2034 smp_mb__before_atomic_dec();
2035 atomic_long_dec(&nfs_access_nr_entries);
2036 smp_mb__after_atomic_dec();
2039 static void nfs_access_free_list(struct list_head *head)
2041 struct nfs_access_entry *cache;
2043 while (!list_empty(head)) {
2044 cache = list_entry(head->next, struct nfs_access_entry, lru);
2045 list_del(&cache->lru);
2046 nfs_access_free_entry(cache);
2050 int nfs_access_cache_shrinker(struct shrinker *shrink,
2051 struct shrink_control *sc)
2054 struct nfs_inode *nfsi, *next;
2055 struct nfs_access_entry *cache;
2056 int nr_to_scan = sc->nr_to_scan;
2057 gfp_t gfp_mask = sc->gfp_mask;
2059 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2060 return (nr_to_scan == 0) ? 0 : -1;
2062 spin_lock(&nfs_access_lru_lock);
2063 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2064 struct inode *inode;
2066 if (nr_to_scan-- == 0)
2068 inode = &nfsi->vfs_inode;
2069 spin_lock(&inode->i_lock);
2070 if (list_empty(&nfsi->access_cache_entry_lru))
2071 goto remove_lru_entry;
2072 cache = list_entry(nfsi->access_cache_entry_lru.next,
2073 struct nfs_access_entry, lru);
2074 list_move(&cache->lru, &head);
2075 rb_erase(&cache->rb_node, &nfsi->access_cache);
2076 if (!list_empty(&nfsi->access_cache_entry_lru))
2077 list_move_tail(&nfsi->access_cache_inode_lru,
2078 &nfs_access_lru_list);
2081 list_del_init(&nfsi->access_cache_inode_lru);
2082 smp_mb__before_clear_bit();
2083 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2084 smp_mb__after_clear_bit();
2086 spin_unlock(&inode->i_lock);
2088 spin_unlock(&nfs_access_lru_lock);
2089 nfs_access_free_list(&head);
2090 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2093 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2095 struct rb_root *root_node = &nfsi->access_cache;
2097 struct nfs_access_entry *entry;
2099 /* Unhook entries from the cache */
2100 while ((n = rb_first(root_node)) != NULL) {
2101 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2102 rb_erase(n, root_node);
2103 list_move(&entry->lru, head);
2105 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2108 void nfs_access_zap_cache(struct inode *inode)
2112 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2114 /* Remove from global LRU init */
2115 spin_lock(&nfs_access_lru_lock);
2116 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2117 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2119 spin_lock(&inode->i_lock);
2120 __nfs_access_zap_cache(NFS_I(inode), &head);
2121 spin_unlock(&inode->i_lock);
2122 spin_unlock(&nfs_access_lru_lock);
2123 nfs_access_free_list(&head);
2126 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2128 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2129 struct nfs_access_entry *entry;
2132 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2134 if (cred < entry->cred)
2136 else if (cred > entry->cred)
2144 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2146 struct nfs_inode *nfsi = NFS_I(inode);
2147 struct nfs_access_entry *cache;
2150 spin_lock(&inode->i_lock);
2151 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2153 cache = nfs_access_search_rbtree(inode, cred);
2156 if (!nfs_have_delegated_attributes(inode) &&
2157 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2159 res->jiffies = cache->jiffies;
2160 res->cred = cache->cred;
2161 res->mask = cache->mask;
2162 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2165 spin_unlock(&inode->i_lock);
2168 rb_erase(&cache->rb_node, &nfsi->access_cache);
2169 list_del(&cache->lru);
2170 spin_unlock(&inode->i_lock);
2171 nfs_access_free_entry(cache);
2174 spin_unlock(&inode->i_lock);
2175 nfs_access_zap_cache(inode);
2179 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2181 struct nfs_inode *nfsi = NFS_I(inode);
2182 struct rb_root *root_node = &nfsi->access_cache;
2183 struct rb_node **p = &root_node->rb_node;
2184 struct rb_node *parent = NULL;
2185 struct nfs_access_entry *entry;
2187 spin_lock(&inode->i_lock);
2188 while (*p != NULL) {
2190 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2192 if (set->cred < entry->cred)
2193 p = &parent->rb_left;
2194 else if (set->cred > entry->cred)
2195 p = &parent->rb_right;
2199 rb_link_node(&set->rb_node, parent, p);
2200 rb_insert_color(&set->rb_node, root_node);
2201 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2202 spin_unlock(&inode->i_lock);
2205 rb_replace_node(parent, &set->rb_node, root_node);
2206 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2207 list_del(&entry->lru);
2208 spin_unlock(&inode->i_lock);
2209 nfs_access_free_entry(entry);
2212 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2214 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2217 RB_CLEAR_NODE(&cache->rb_node);
2218 cache->jiffies = set->jiffies;
2219 cache->cred = get_rpccred(set->cred);
2220 cache->mask = set->mask;
2222 nfs_access_add_rbtree(inode, cache);
2224 /* Update accounting */
2225 smp_mb__before_atomic_inc();
2226 atomic_long_inc(&nfs_access_nr_entries);
2227 smp_mb__after_atomic_inc();
2229 /* Add inode to global LRU list */
2230 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2231 spin_lock(&nfs_access_lru_lock);
2232 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2233 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2234 &nfs_access_lru_list);
2235 spin_unlock(&nfs_access_lru_lock);
2239 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2241 struct nfs_access_entry cache;
2244 status = nfs_access_get_cached(inode, cred, &cache);
2248 /* Be clever: ask server to check for all possible rights */
2249 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2251 cache.jiffies = jiffies;
2252 status = NFS_PROTO(inode)->access(inode, &cache);
2254 if (status == -ESTALE) {
2255 nfs_zap_caches(inode);
2256 if (!S_ISDIR(inode->i_mode))
2257 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2261 nfs_access_add_cache(inode, &cache);
2263 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2268 static int nfs_open_permission_mask(int openflags)
2272 if (openflags & FMODE_READ)
2274 if (openflags & FMODE_WRITE)
2276 if (openflags & FMODE_EXEC)
2281 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2283 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2286 int nfs_permission(struct inode *inode, int mask, unsigned int flags)
2288 struct rpc_cred *cred;
2291 if (flags & IPERM_FLAG_RCU)
2294 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2296 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2298 /* Is this sys_access() ? */
2299 if (mask & (MAY_ACCESS | MAY_CHDIR))
2302 switch (inode->i_mode & S_IFMT) {
2306 /* NFSv4 has atomic_open... */
2307 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2308 && (mask & MAY_OPEN)
2309 && !(mask & MAY_EXEC))
2314 * Optimize away all write operations, since the server
2315 * will check permissions when we perform the op.
2317 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2322 if (!NFS_PROTO(inode)->access)
2325 cred = rpc_lookup_cred();
2326 if (!IS_ERR(cred)) {
2327 res = nfs_do_access(inode, cred, mask);
2330 res = PTR_ERR(cred);
2332 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2335 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2336 inode->i_sb->s_id, inode->i_ino, mask, res);
2339 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2341 res = generic_permission(inode, mask, flags, NULL);
2347 * version-control: t
2348 * kept-new-versions: 5