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 *, umode_t, struct nameidata *);
51 static int nfs_mkdir(struct inode *, struct dentry *, umode_t);
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 *, umode_t, dev_t);
57 static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59 static int nfs_fsync_dir(struct file *, loff_t, loff_t, 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, umode_t 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);
264 for (i = 0; i < array->size; i++)
265 kfree(array->array[i].string.name);
266 kunmap_atomic(array);
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 "The file: %s has duplicate cookie %llu\n",
369 desc->file->f_dentry->d_parent->d_name.name,
370 desc->file->f_dentry->d_name.name,
371 array->array[i].string.name,
377 ctx->dup_cookie = *desc->dir_cookie;
380 desc->file->f_pos = new_pos;
381 desc->cache_entry_index = i;
385 if (array->eof_index >= 0) {
386 status = -EBADCOOKIE;
387 if (*desc->dir_cookie == array->last_cookie)
395 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
397 struct nfs_cache_array *array;
400 array = nfs_readdir_get_array(desc->page);
402 status = PTR_ERR(array);
406 if (*desc->dir_cookie == 0)
407 status = nfs_readdir_search_for_pos(array, desc);
409 status = nfs_readdir_search_for_cookie(array, desc);
411 if (status == -EAGAIN) {
412 desc->last_cookie = array->last_cookie;
413 desc->current_index += array->size;
416 nfs_readdir_release_array(desc->page);
421 /* Fill a page with xdr information before transferring to the cache page */
423 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
424 struct nfs_entry *entry, struct file *file, struct inode *inode)
426 struct nfs_open_dir_context *ctx = file->private_data;
427 struct rpc_cred *cred = ctx->cred;
428 unsigned long timestamp, gencount;
433 gencount = nfs_inc_attr_generation_counter();
434 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
435 NFS_SERVER(inode)->dtsize, desc->plus);
437 /* We requested READDIRPLUS, but the server doesn't grok it */
438 if (error == -ENOTSUPP && desc->plus) {
439 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
440 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
446 desc->timestamp = timestamp;
447 desc->gencount = gencount;
452 static int xdr_decode(nfs_readdir_descriptor_t *desc,
453 struct nfs_entry *entry, struct xdr_stream *xdr)
457 error = desc->decode(xdr, entry, desc->plus);
460 entry->fattr->time_start = desc->timestamp;
461 entry->fattr->gencount = desc->gencount;
466 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
468 if (dentry->d_inode == NULL)
470 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
478 bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
480 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
482 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
484 if (filp->f_pos == 0)
490 * This function is called by the lookup code to request the use of
491 * readdirplus to accelerate any future lookups in the same
495 void nfs_advise_use_readdirplus(struct inode *dir)
497 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
501 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
503 struct qstr filename = {
507 struct dentry *dentry;
508 struct dentry *alias;
509 struct inode *dir = parent->d_inode;
512 if (filename.name[0] == '.') {
513 if (filename.len == 1)
515 if (filename.len == 2 && filename.name[1] == '.')
518 filename.hash = full_name_hash(filename.name, filename.len);
520 dentry = d_lookup(parent, &filename);
521 if (dentry != NULL) {
522 if (nfs_same_file(dentry, entry)) {
523 nfs_refresh_inode(dentry->d_inode, entry->fattr);
531 dentry = d_alloc(parent, &filename);
535 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
539 alias = d_materialise_unique(dentry, inode);
543 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
546 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
552 /* Perform conversion from xdr to cache array */
554 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
555 struct page **xdr_pages, struct page *page, unsigned int buflen)
557 struct xdr_stream stream;
559 struct page *scratch;
560 struct nfs_cache_array *array;
561 unsigned int count = 0;
564 scratch = alloc_page(GFP_KERNEL);
568 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
569 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
572 status = xdr_decode(desc, entry, &stream);
574 if (status == -EAGAIN)
582 nfs_prime_dcache(desc->file->f_path.dentry, entry);
584 status = nfs_readdir_add_to_array(entry, page);
587 } while (!entry->eof);
589 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
590 array = nfs_readdir_get_array(page);
591 if (!IS_ERR(array)) {
592 array->eof_index = array->size;
594 nfs_readdir_release_array(page);
596 status = PTR_ERR(array);
604 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
607 for (i = 0; i < npages; i++)
612 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
615 nfs_readdir_free_pagearray(pages, npages);
619 * nfs_readdir_large_page will allocate pages that must be freed with a call
620 * to nfs_readdir_free_large_page
623 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
627 for (i = 0; i < npages; i++) {
628 struct page *page = alloc_page(GFP_KERNEL);
636 nfs_readdir_free_pagearray(pages, i);
641 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
643 struct page *pages[NFS_MAX_READDIR_PAGES];
644 void *pages_ptr = NULL;
645 struct nfs_entry entry;
646 struct file *file = desc->file;
647 struct nfs_cache_array *array;
648 int status = -ENOMEM;
649 unsigned int array_size = ARRAY_SIZE(pages);
651 entry.prev_cookie = 0;
652 entry.cookie = desc->last_cookie;
654 entry.fh = nfs_alloc_fhandle();
655 entry.fattr = nfs_alloc_fattr();
656 entry.server = NFS_SERVER(inode);
657 if (entry.fh == NULL || entry.fattr == NULL)
660 array = nfs_readdir_get_array(page);
662 status = PTR_ERR(array);
665 memset(array, 0, sizeof(struct nfs_cache_array));
666 array->eof_index = -1;
668 status = nfs_readdir_large_page(pages, array_size);
670 goto out_release_array;
673 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
678 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
680 if (status == -ENOSPC)
684 } while (array->eof_index < 0);
686 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
688 nfs_readdir_release_array(page);
690 nfs_free_fattr(entry.fattr);
691 nfs_free_fhandle(entry.fh);
696 * Now we cache directories properly, by converting xdr information
697 * to an array that can be used for lookups later. This results in
698 * fewer cache pages, since we can store more information on each page.
699 * We only need to convert from xdr once so future lookups are much simpler
702 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
704 struct inode *inode = desc->file->f_path.dentry->d_inode;
707 ret = nfs_readdir_xdr_to_array(desc, page, inode);
710 SetPageUptodate(page);
712 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
713 /* Should never happen */
714 nfs_zap_mapping(inode, inode->i_mapping);
724 void cache_page_release(nfs_readdir_descriptor_t *desc)
726 if (!desc->page->mapping)
727 nfs_readdir_clear_array(desc->page);
728 page_cache_release(desc->page);
733 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
735 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
736 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
740 * Returns 0 if desc->dir_cookie was found on page desc->page_index
743 int find_cache_page(nfs_readdir_descriptor_t *desc)
747 desc->page = get_cache_page(desc);
748 if (IS_ERR(desc->page))
749 return PTR_ERR(desc->page);
751 res = nfs_readdir_search_array(desc);
753 cache_page_release(desc);
757 /* Search for desc->dir_cookie from the beginning of the page cache */
759 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
763 if (desc->page_index == 0) {
764 desc->current_index = 0;
765 desc->last_cookie = 0;
768 res = find_cache_page(desc);
769 } while (res == -EAGAIN);
774 * Once we've found the start of the dirent within a page: fill 'er up...
777 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
780 struct file *file = desc->file;
783 struct nfs_cache_array *array = NULL;
784 struct nfs_open_dir_context *ctx = file->private_data;
786 array = nfs_readdir_get_array(desc->page);
788 res = PTR_ERR(array);
792 for (i = desc->cache_entry_index; i < array->size; i++) {
793 struct nfs_cache_array_entry *ent;
795 ent = &array->array[i];
796 if (filldir(dirent, ent->string.name, ent->string.len,
797 file->f_pos, nfs_compat_user_ino64(ent->ino),
803 if (i < (array->size-1))
804 *desc->dir_cookie = array->array[i+1].cookie;
806 *desc->dir_cookie = array->last_cookie;
810 if (array->eof_index >= 0)
813 nfs_readdir_release_array(desc->page);
815 cache_page_release(desc);
816 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
817 (unsigned long long)*desc->dir_cookie, res);
822 * If we cannot find a cookie in our cache, we suspect that this is
823 * because it points to a deleted file, so we ask the server to return
824 * whatever it thinks is the next entry. We then feed this to filldir.
825 * If all goes well, we should then be able to find our way round the
826 * cache on the next call to readdir_search_pagecache();
828 * NOTE: we cannot add the anonymous page to the pagecache because
829 * the data it contains might not be page aligned. Besides,
830 * we should already have a complete representation of the
831 * directory in the page cache by the time we get here.
834 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
837 struct page *page = NULL;
839 struct inode *inode = desc->file->f_path.dentry->d_inode;
840 struct nfs_open_dir_context *ctx = desc->file->private_data;
842 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
843 (unsigned long long)*desc->dir_cookie);
845 page = alloc_page(GFP_HIGHUSER);
851 desc->page_index = 0;
852 desc->last_cookie = *desc->dir_cookie;
856 status = nfs_readdir_xdr_to_array(desc, page, inode);
860 status = nfs_do_filldir(desc, dirent, filldir);
863 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
867 cache_page_release(desc);
871 /* The file offset position represents the dirent entry number. A
872 last cookie cache takes care of the common case of reading the
875 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
877 struct dentry *dentry = filp->f_path.dentry;
878 struct inode *inode = dentry->d_inode;
879 nfs_readdir_descriptor_t my_desc,
881 struct nfs_open_dir_context *dir_ctx = filp->private_data;
884 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
885 dentry->d_parent->d_name.name, dentry->d_name.name,
886 (long long)filp->f_pos);
887 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
890 * filp->f_pos points to the dirent entry number.
891 * *desc->dir_cookie has the cookie for the next entry. We have
892 * to either find the entry with the appropriate number or
893 * revalidate the cookie.
895 memset(desc, 0, sizeof(*desc));
898 desc->dir_cookie = &dir_ctx->dir_cookie;
899 desc->decode = NFS_PROTO(inode)->decode_dirent;
900 desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
902 nfs_block_sillyrename(dentry);
903 res = nfs_revalidate_mapping(inode, filp->f_mapping);
908 res = readdir_search_pagecache(desc);
910 if (res == -EBADCOOKIE) {
912 /* This means either end of directory */
913 if (*desc->dir_cookie && desc->eof == 0) {
914 /* Or that the server has 'lost' a cookie */
915 res = uncached_readdir(desc, dirent, filldir);
921 if (res == -ETOOSMALL && desc->plus) {
922 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
923 nfs_zap_caches(inode);
924 desc->page_index = 0;
932 res = nfs_do_filldir(desc, dirent, filldir);
935 } while (!desc->eof);
937 nfs_unblock_sillyrename(dentry);
940 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
941 dentry->d_parent->d_name.name, dentry->d_name.name,
946 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
948 struct dentry *dentry = filp->f_path.dentry;
949 struct inode *inode = dentry->d_inode;
950 struct nfs_open_dir_context *dir_ctx = filp->private_data;
952 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
953 dentry->d_parent->d_name.name,
957 mutex_lock(&inode->i_mutex);
960 offset += filp->f_pos;
968 if (offset != filp->f_pos) {
969 filp->f_pos = offset;
970 dir_ctx->dir_cookie = 0;
974 mutex_unlock(&inode->i_mutex);
979 * All directory operations under NFS are synchronous, so fsync()
980 * is a dummy operation.
982 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
985 struct dentry *dentry = filp->f_path.dentry;
986 struct inode *inode = dentry->d_inode;
988 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
989 dentry->d_parent->d_name.name, dentry->d_name.name,
992 mutex_lock(&inode->i_mutex);
993 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
994 mutex_unlock(&inode->i_mutex);
999 * nfs_force_lookup_revalidate - Mark the directory as having changed
1000 * @dir - pointer to directory inode
1002 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1003 * full lookup on all child dentries of 'dir' whenever a change occurs
1004 * on the server that might have invalidated our dcache.
1006 * The caller should be holding dir->i_lock
1008 void nfs_force_lookup_revalidate(struct inode *dir)
1010 NFS_I(dir)->cache_change_attribute++;
1014 * A check for whether or not the parent directory has changed.
1015 * In the case it has, we assume that the dentries are untrustworthy
1016 * and may need to be looked up again.
1018 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1020 if (IS_ROOT(dentry))
1022 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1024 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1026 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1027 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1029 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1035 * Return the intent data that applies to this particular path component
1037 * Note that the current set of intents only apply to the very last
1038 * component of the path and none of them is set before that last
1041 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1044 return nd->flags & mask;
1048 * Use intent information to check whether or not we're going to do
1049 * an O_EXCL create using this path component.
1051 static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1053 if (NFS_PROTO(dir)->version == 2)
1055 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1059 * Inode and filehandle revalidation for lookups.
1061 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1062 * or if the intent information indicates that we're about to open this
1063 * particular file and the "nocto" mount flag is not set.
1067 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1069 struct nfs_server *server = NFS_SERVER(inode);
1071 if (IS_AUTOMOUNT(inode))
1074 /* VFS wants an on-the-wire revalidation */
1075 if (nd->flags & LOOKUP_REVAL)
1077 /* This is an open(2) */
1078 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1079 !(server->flags & NFS_MOUNT_NOCTO) &&
1080 (S_ISREG(inode->i_mode) ||
1081 S_ISDIR(inode->i_mode)))
1085 return nfs_revalidate_inode(server, inode);
1087 return __nfs_revalidate_inode(server, inode);
1091 * We judge how long we want to trust negative
1092 * dentries by looking at the parent inode mtime.
1094 * If parent mtime has changed, we revalidate, else we wait for a
1095 * period corresponding to the parent's attribute cache timeout value.
1098 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1099 struct nameidata *nd)
1101 /* Don't revalidate a negative dentry if we're creating a new file */
1102 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1104 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1106 return !nfs_check_verifier(dir, dentry);
1110 * This is called every time the dcache has a lookup hit,
1111 * and we should check whether we can really trust that
1114 * NOTE! The hit can be a negative hit too, don't assume
1117 * If the parent directory is seen to have changed, we throw out the
1118 * cached dentry and do a new lookup.
1120 static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1123 struct inode *inode;
1124 struct dentry *parent;
1125 struct nfs_fh *fhandle = NULL;
1126 struct nfs_fattr *fattr = NULL;
1129 if (nd->flags & LOOKUP_RCU)
1132 parent = dget_parent(dentry);
1133 dir = parent->d_inode;
1134 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1135 inode = dentry->d_inode;
1138 if (nfs_neg_need_reval(dir, dentry, nd))
1140 goto out_valid_noent;
1143 if (is_bad_inode(inode)) {
1144 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1145 __func__, dentry->d_parent->d_name.name,
1146 dentry->d_name.name);
1150 if (nfs_have_delegation(inode, FMODE_READ))
1151 goto out_set_verifier;
1153 /* Force a full look up iff the parent directory has changed */
1154 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1155 if (nfs_lookup_verify_inode(inode, nd))
1156 goto out_zap_parent;
1160 if (NFS_STALE(inode))
1164 fhandle = nfs_alloc_fhandle();
1165 fattr = nfs_alloc_fattr();
1166 if (fhandle == NULL || fattr == NULL)
1169 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1172 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1174 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1177 nfs_free_fattr(fattr);
1178 nfs_free_fhandle(fhandle);
1180 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1182 /* Success: notify readdir to use READDIRPLUS */
1183 nfs_advise_use_readdirplus(dir);
1186 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1187 __func__, dentry->d_parent->d_name.name,
1188 dentry->d_name.name);
1191 nfs_zap_caches(dir);
1193 nfs_mark_for_revalidate(dir);
1194 if (inode && S_ISDIR(inode->i_mode)) {
1195 /* Purge readdir caches. */
1196 nfs_zap_caches(inode);
1197 /* If we have submounts, don't unhash ! */
1198 if (have_submounts(dentry))
1200 if (dentry->d_flags & DCACHE_DISCONNECTED)
1202 shrink_dcache_parent(dentry);
1205 nfs_free_fattr(fattr);
1206 nfs_free_fhandle(fhandle);
1208 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1209 __func__, dentry->d_parent->d_name.name,
1210 dentry->d_name.name);
1213 nfs_free_fattr(fattr);
1214 nfs_free_fhandle(fhandle);
1216 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1217 __func__, dentry->d_parent->d_name.name,
1218 dentry->d_name.name, error);
1223 * This is called from dput() when d_count is going to 0.
1225 static int nfs_dentry_delete(const struct dentry *dentry)
1227 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1228 dentry->d_parent->d_name.name, dentry->d_name.name,
1231 /* Unhash any dentry with a stale inode */
1232 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1235 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1236 /* Unhash it, so that ->d_iput() would be called */
1239 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1240 /* Unhash it, so that ancestors of killed async unlink
1241 * files will be cleaned up during umount */
1248 static void nfs_drop_nlink(struct inode *inode)
1250 spin_lock(&inode->i_lock);
1251 if (inode->i_nlink > 0)
1253 spin_unlock(&inode->i_lock);
1257 * Called when the dentry loses inode.
1258 * We use it to clean up silly-renamed files.
1260 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1262 if (S_ISDIR(inode->i_mode))
1263 /* drop any readdir cache as it could easily be old */
1264 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1266 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1268 nfs_complete_unlink(dentry, inode);
1273 static void nfs_d_release(struct dentry *dentry)
1275 /* free cached devname value, if it survived that far */
1276 if (unlikely(dentry->d_fsdata)) {
1277 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1280 kfree(dentry->d_fsdata);
1284 const struct dentry_operations nfs_dentry_operations = {
1285 .d_revalidate = nfs_lookup_revalidate,
1286 .d_delete = nfs_dentry_delete,
1287 .d_iput = nfs_dentry_iput,
1288 .d_automount = nfs_d_automount,
1289 .d_release = nfs_d_release,
1292 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1295 struct dentry *parent;
1296 struct inode *inode = NULL;
1297 struct nfs_fh *fhandle = NULL;
1298 struct nfs_fattr *fattr = NULL;
1301 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1302 dentry->d_parent->d_name.name, dentry->d_name.name);
1303 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1305 res = ERR_PTR(-ENAMETOOLONG);
1306 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1310 * If we're doing an exclusive create, optimize away the lookup
1311 * but don't hash the dentry.
1313 if (nfs_is_exclusive_create(dir, nd)) {
1314 d_instantiate(dentry, NULL);
1319 res = ERR_PTR(-ENOMEM);
1320 fhandle = nfs_alloc_fhandle();
1321 fattr = nfs_alloc_fattr();
1322 if (fhandle == NULL || fattr == NULL)
1325 parent = dentry->d_parent;
1326 /* Protect against concurrent sillydeletes */
1327 nfs_block_sillyrename(parent);
1328 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1329 if (error == -ENOENT)
1332 res = ERR_PTR(error);
1333 goto out_unblock_sillyrename;
1335 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1336 res = ERR_CAST(inode);
1338 goto out_unblock_sillyrename;
1340 /* Success: notify readdir to use READDIRPLUS */
1341 nfs_advise_use_readdirplus(dir);
1344 res = d_materialise_unique(dentry, inode);
1347 goto out_unblock_sillyrename;
1350 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1351 out_unblock_sillyrename:
1352 nfs_unblock_sillyrename(parent);
1354 nfs_free_fattr(fattr);
1355 nfs_free_fhandle(fhandle);
1359 #ifdef CONFIG_NFS_V4
1360 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1362 const struct dentry_operations nfs4_dentry_operations = {
1363 .d_revalidate = nfs_open_revalidate,
1364 .d_delete = nfs_dentry_delete,
1365 .d_iput = nfs_dentry_iput,
1366 .d_automount = nfs_d_automount,
1367 .d_release = nfs_d_release,
1371 * Use intent information to determine whether we need to substitute
1372 * the NFSv4-style stateful OPEN for the LOOKUP call
1374 static int is_atomic_open(struct nameidata *nd)
1376 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1378 /* NFS does not (yet) have a stateful open for directories */
1379 if (nd->flags & LOOKUP_DIRECTORY)
1381 /* Are we trying to write to a read only partition? */
1382 if (__mnt_is_readonly(nd->path.mnt) &&
1383 (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1388 static fmode_t flags_to_mode(int flags)
1390 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1391 if ((flags & O_ACCMODE) != O_WRONLY)
1393 if ((flags & O_ACCMODE) != O_RDONLY)
1398 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1400 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1403 static int do_open(struct inode *inode, struct file *filp)
1405 nfs_fscache_set_inode_cookie(inode, filp);
1409 static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1414 /* If the open_intent is for execute, we have an extra check to make */
1415 if (ctx->mode & FMODE_EXEC) {
1416 ret = nfs_may_open(ctx->dentry->d_inode,
1418 nd->intent.open.flags);
1422 filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1424 ret = PTR_ERR(filp);
1426 nfs_file_set_open_context(filp, ctx);
1428 put_nfs_open_context(ctx);
1432 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1434 struct nfs_open_context *ctx;
1436 struct dentry *res = NULL;
1437 struct inode *inode;
1441 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1442 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1444 /* Check that we are indeed trying to open this file */
1445 if (!is_atomic_open(nd))
1448 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1449 res = ERR_PTR(-ENAMETOOLONG);
1453 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1455 if (nd->flags & LOOKUP_EXCL) {
1456 d_instantiate(dentry, NULL);
1460 open_flags = nd->intent.open.flags;
1461 attr.ia_valid = ATTR_OPEN;
1463 ctx = create_nfs_open_context(dentry, open_flags);
1464 res = ERR_CAST(ctx);
1468 if (nd->flags & LOOKUP_CREATE) {
1469 attr.ia_mode = nd->intent.open.create_mode;
1470 attr.ia_valid |= ATTR_MODE;
1471 attr.ia_mode &= ~current_umask();
1473 open_flags &= ~(O_EXCL | O_CREAT);
1475 if (open_flags & O_TRUNC) {
1476 attr.ia_valid |= ATTR_SIZE;
1480 /* Open the file on the server */
1481 nfs_block_sillyrename(dentry->d_parent);
1482 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1483 if (IS_ERR(inode)) {
1484 nfs_unblock_sillyrename(dentry->d_parent);
1485 put_nfs_open_context(ctx);
1486 switch (PTR_ERR(inode)) {
1487 /* Make a negative dentry */
1489 d_add(dentry, NULL);
1492 /* This turned out not to be a regular file */
1497 if (!(nd->intent.open.flags & O_NOFOLLOW))
1501 res = ERR_CAST(inode);
1505 res = d_add_unique(dentry, inode);
1506 nfs_unblock_sillyrename(dentry->d_parent);
1509 ctx->dentry = dget(res);
1512 err = nfs_intent_set_file(nd, ctx);
1516 return ERR_PTR(err);
1519 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1522 return nfs_lookup(dir, dentry, nd);
1525 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1527 struct dentry *parent = NULL;
1528 struct inode *inode;
1530 struct nfs_open_context *ctx;
1532 int openflags, ret = 0;
1534 if (nd->flags & LOOKUP_RCU)
1537 inode = dentry->d_inode;
1538 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1541 parent = dget_parent(dentry);
1542 dir = parent->d_inode;
1544 /* We can't create new files in nfs_open_revalidate(), so we
1545 * optimize away revalidation of negative dentries.
1547 if (inode == NULL) {
1548 if (!nfs_neg_need_reval(dir, dentry, nd))
1553 /* NFS only supports OPEN on regular files */
1554 if (!S_ISREG(inode->i_mode))
1556 openflags = nd->intent.open.flags;
1557 /* We cannot do exclusive creation on a positive dentry */
1558 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1560 /* We can't create new files here */
1561 openflags &= ~(O_CREAT|O_EXCL);
1563 ctx = create_nfs_open_context(dentry, openflags);
1568 attr.ia_valid = ATTR_OPEN;
1569 if (openflags & O_TRUNC) {
1570 attr.ia_valid |= ATTR_SIZE;
1576 * Note: we're not holding inode->i_mutex and so may be racing with
1577 * operations that change the directory. We therefore save the
1578 * change attribute *before* we do the RPC call.
1580 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr);
1581 if (IS_ERR(inode)) {
1582 ret = PTR_ERR(inode);
1595 if (inode != dentry->d_inode)
1598 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1599 ret = nfs_intent_set_file(nd, ctx);
1609 put_nfs_open_context(ctx);
1615 return nfs_lookup_revalidate(dentry, nd);
1618 static int nfs_open_create(struct inode *dir, struct dentry *dentry,
1619 umode_t mode, struct nameidata *nd)
1621 struct nfs_open_context *ctx = NULL;
1624 int open_flags = O_CREAT|O_EXCL;
1626 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1627 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1629 attr.ia_mode = mode;
1630 attr.ia_valid = ATTR_MODE;
1633 open_flags = nd->intent.open.flags;
1635 ctx = create_nfs_open_context(dentry, open_flags);
1636 error = PTR_ERR(ctx);
1640 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1644 error = nfs_intent_set_file(nd, ctx);
1648 put_nfs_open_context(ctx);
1652 put_nfs_open_context(ctx);
1659 #endif /* CONFIG_NFSV4 */
1662 * Code common to create, mkdir, and mknod.
1664 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1665 struct nfs_fattr *fattr)
1667 struct dentry *parent = dget_parent(dentry);
1668 struct inode *dir = parent->d_inode;
1669 struct inode *inode;
1670 int error = -EACCES;
1674 /* We may have been initialized further down */
1675 if (dentry->d_inode)
1677 if (fhandle->size == 0) {
1678 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1682 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1683 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1684 struct nfs_server *server = NFS_SB(dentry->d_sb);
1685 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1689 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1690 error = PTR_ERR(inode);
1693 d_add(dentry, inode);
1698 nfs_mark_for_revalidate(dir);
1704 * Following a failed create operation, we drop the dentry rather
1705 * than retain a negative dentry. This avoids a problem in the event
1706 * that the operation succeeded on the server, but an error in the
1707 * reply path made it appear to have failed.
1709 static int nfs_create(struct inode *dir, struct dentry *dentry,
1710 umode_t mode, struct nameidata *nd)
1714 int open_flags = O_CREAT|O_EXCL;
1716 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1717 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1719 attr.ia_mode = mode;
1720 attr.ia_valid = ATTR_MODE;
1723 open_flags = nd->intent.open.flags;
1725 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1735 * See comments for nfs_proc_create regarding failed operations.
1738 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1743 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1744 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1746 if (!new_valid_dev(rdev))
1749 attr.ia_mode = mode;
1750 attr.ia_valid = ATTR_MODE;
1752 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1762 * See comments for nfs_proc_create regarding failed operations.
1764 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1769 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1770 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1772 attr.ia_valid = ATTR_MODE;
1773 attr.ia_mode = mode | S_IFDIR;
1775 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1784 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1786 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1790 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1794 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1795 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1797 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1798 /* Ensure the VFS deletes this inode */
1799 if (error == 0 && dentry->d_inode != NULL)
1800 clear_nlink(dentry->d_inode);
1801 else if (error == -ENOENT)
1802 nfs_dentry_handle_enoent(dentry);
1808 * Remove a file after making sure there are no pending writes,
1809 * and after checking that the file has only one user.
1811 * We invalidate the attribute cache and free the inode prior to the operation
1812 * to avoid possible races if the server reuses the inode.
1814 static int nfs_safe_remove(struct dentry *dentry)
1816 struct inode *dir = dentry->d_parent->d_inode;
1817 struct inode *inode = dentry->d_inode;
1820 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1821 dentry->d_parent->d_name.name, dentry->d_name.name);
1823 /* If the dentry was sillyrenamed, we simply call d_delete() */
1824 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1829 if (inode != NULL) {
1830 nfs_inode_return_delegation(inode);
1831 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1832 /* The VFS may want to delete this inode */
1834 nfs_drop_nlink(inode);
1835 nfs_mark_for_revalidate(inode);
1837 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1838 if (error == -ENOENT)
1839 nfs_dentry_handle_enoent(dentry);
1844 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1845 * belongs to an active ".nfs..." file and we return -EBUSY.
1847 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1849 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1852 int need_rehash = 0;
1854 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1855 dir->i_ino, dentry->d_name.name);
1857 spin_lock(&dentry->d_lock);
1858 if (dentry->d_count > 1) {
1859 spin_unlock(&dentry->d_lock);
1860 /* Start asynchronous writeout of the inode */
1861 write_inode_now(dentry->d_inode, 0);
1862 error = nfs_sillyrename(dir, dentry);
1865 if (!d_unhashed(dentry)) {
1869 spin_unlock(&dentry->d_lock);
1870 error = nfs_safe_remove(dentry);
1871 if (!error || error == -ENOENT) {
1872 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1873 } else if (need_rehash)
1879 * To create a symbolic link, most file systems instantiate a new inode,
1880 * add a page to it containing the path, then write it out to the disk
1881 * using prepare_write/commit_write.
1883 * Unfortunately the NFS client can't create the in-core inode first
1884 * because it needs a file handle to create an in-core inode (see
1885 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1886 * symlink request has completed on the server.
1888 * So instead we allocate a raw page, copy the symname into it, then do
1889 * the SYMLINK request with the page as the buffer. If it succeeds, we
1890 * now have a new file handle and can instantiate an in-core NFS inode
1891 * and move the raw page into its mapping.
1893 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1895 struct pagevec lru_pvec;
1899 unsigned int pathlen = strlen(symname);
1902 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1903 dir->i_ino, dentry->d_name.name, symname);
1905 if (pathlen > PAGE_SIZE)
1906 return -ENAMETOOLONG;
1908 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1909 attr.ia_valid = ATTR_MODE;
1911 page = alloc_page(GFP_HIGHUSER);
1915 kaddr = kmap_atomic(page);
1916 memcpy(kaddr, symname, pathlen);
1917 if (pathlen < PAGE_SIZE)
1918 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1919 kunmap_atomic(kaddr);
1921 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1923 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1924 dir->i_sb->s_id, dir->i_ino,
1925 dentry->d_name.name, symname, error);
1932 * No big deal if we can't add this page to the page cache here.
1933 * READLINK will get the missing page from the server if needed.
1935 pagevec_init(&lru_pvec, 0);
1936 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1938 pagevec_add(&lru_pvec, page);
1939 pagevec_lru_add_file(&lru_pvec);
1940 SetPageUptodate(page);
1949 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1951 struct inode *inode = old_dentry->d_inode;
1954 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1955 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1956 dentry->d_parent->d_name.name, dentry->d_name.name);
1958 nfs_inode_return_delegation(inode);
1961 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1964 d_add(dentry, inode);
1971 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1972 * different file handle for the same inode after a rename (e.g. when
1973 * moving to a different directory). A fail-safe method to do so would
1974 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1975 * rename the old file using the sillyrename stuff. This way, the original
1976 * file in old_dir will go away when the last process iput()s the inode.
1980 * It actually works quite well. One needs to have the possibility for
1981 * at least one ".nfs..." file in each directory the file ever gets
1982 * moved or linked to which happens automagically with the new
1983 * implementation that only depends on the dcache stuff instead of
1984 * using the inode layer
1986 * Unfortunately, things are a little more complicated than indicated
1987 * above. For a cross-directory move, we want to make sure we can get
1988 * rid of the old inode after the operation. This means there must be
1989 * no pending writes (if it's a file), and the use count must be 1.
1990 * If these conditions are met, we can drop the dentries before doing
1993 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1994 struct inode *new_dir, struct dentry *new_dentry)
1996 struct inode *old_inode = old_dentry->d_inode;
1997 struct inode *new_inode = new_dentry->d_inode;
1998 struct dentry *dentry = NULL, *rehash = NULL;
2001 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
2002 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
2003 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
2004 new_dentry->d_count);
2007 * For non-directories, check whether the target is busy and if so,
2008 * make a copy of the dentry and then do a silly-rename. If the
2009 * silly-rename succeeds, the copied dentry is hashed and becomes
2012 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2014 * To prevent any new references to the target during the
2015 * rename, we unhash the dentry in advance.
2017 if (!d_unhashed(new_dentry)) {
2019 rehash = new_dentry;
2022 if (new_dentry->d_count > 2) {
2025 /* copy the target dentry's name */
2026 dentry = d_alloc(new_dentry->d_parent,
2027 &new_dentry->d_name);
2031 /* silly-rename the existing target ... */
2032 err = nfs_sillyrename(new_dir, new_dentry);
2036 new_dentry = dentry;
2042 nfs_inode_return_delegation(old_inode);
2043 if (new_inode != NULL)
2044 nfs_inode_return_delegation(new_inode);
2046 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
2047 new_dir, &new_dentry->d_name);
2048 nfs_mark_for_revalidate(old_inode);
2053 if (new_inode != NULL)
2054 nfs_drop_nlink(new_inode);
2055 d_move(old_dentry, new_dentry);
2056 nfs_set_verifier(new_dentry,
2057 nfs_save_change_attribute(new_dir));
2058 } else if (error == -ENOENT)
2059 nfs_dentry_handle_enoent(old_dentry);
2061 /* new dentry created? */
2067 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2068 static LIST_HEAD(nfs_access_lru_list);
2069 static atomic_long_t nfs_access_nr_entries;
2071 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2073 put_rpccred(entry->cred);
2075 smp_mb__before_atomic_dec();
2076 atomic_long_dec(&nfs_access_nr_entries);
2077 smp_mb__after_atomic_dec();
2080 static void nfs_access_free_list(struct list_head *head)
2082 struct nfs_access_entry *cache;
2084 while (!list_empty(head)) {
2085 cache = list_entry(head->next, struct nfs_access_entry, lru);
2086 list_del(&cache->lru);
2087 nfs_access_free_entry(cache);
2091 int nfs_access_cache_shrinker(struct shrinker *shrink,
2092 struct shrink_control *sc)
2095 struct nfs_inode *nfsi, *next;
2096 struct nfs_access_entry *cache;
2097 int nr_to_scan = sc->nr_to_scan;
2098 gfp_t gfp_mask = sc->gfp_mask;
2100 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2101 return (nr_to_scan == 0) ? 0 : -1;
2103 spin_lock(&nfs_access_lru_lock);
2104 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2105 struct inode *inode;
2107 if (nr_to_scan-- == 0)
2109 inode = &nfsi->vfs_inode;
2110 spin_lock(&inode->i_lock);
2111 if (list_empty(&nfsi->access_cache_entry_lru))
2112 goto remove_lru_entry;
2113 cache = list_entry(nfsi->access_cache_entry_lru.next,
2114 struct nfs_access_entry, lru);
2115 list_move(&cache->lru, &head);
2116 rb_erase(&cache->rb_node, &nfsi->access_cache);
2117 if (!list_empty(&nfsi->access_cache_entry_lru))
2118 list_move_tail(&nfsi->access_cache_inode_lru,
2119 &nfs_access_lru_list);
2122 list_del_init(&nfsi->access_cache_inode_lru);
2123 smp_mb__before_clear_bit();
2124 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2125 smp_mb__after_clear_bit();
2127 spin_unlock(&inode->i_lock);
2129 spin_unlock(&nfs_access_lru_lock);
2130 nfs_access_free_list(&head);
2131 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2134 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2136 struct rb_root *root_node = &nfsi->access_cache;
2138 struct nfs_access_entry *entry;
2140 /* Unhook entries from the cache */
2141 while ((n = rb_first(root_node)) != NULL) {
2142 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2143 rb_erase(n, root_node);
2144 list_move(&entry->lru, head);
2146 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2149 void nfs_access_zap_cache(struct inode *inode)
2153 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2155 /* Remove from global LRU init */
2156 spin_lock(&nfs_access_lru_lock);
2157 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2158 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2160 spin_lock(&inode->i_lock);
2161 __nfs_access_zap_cache(NFS_I(inode), &head);
2162 spin_unlock(&inode->i_lock);
2163 spin_unlock(&nfs_access_lru_lock);
2164 nfs_access_free_list(&head);
2167 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2169 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2170 struct nfs_access_entry *entry;
2173 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2175 if (cred < entry->cred)
2177 else if (cred > entry->cred)
2185 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2187 struct nfs_inode *nfsi = NFS_I(inode);
2188 struct nfs_access_entry *cache;
2191 spin_lock(&inode->i_lock);
2192 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2194 cache = nfs_access_search_rbtree(inode, cred);
2197 if (!nfs_have_delegated_attributes(inode) &&
2198 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2200 res->jiffies = cache->jiffies;
2201 res->cred = cache->cred;
2202 res->mask = cache->mask;
2203 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2206 spin_unlock(&inode->i_lock);
2209 rb_erase(&cache->rb_node, &nfsi->access_cache);
2210 list_del(&cache->lru);
2211 spin_unlock(&inode->i_lock);
2212 nfs_access_free_entry(cache);
2215 spin_unlock(&inode->i_lock);
2216 nfs_access_zap_cache(inode);
2220 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2222 struct nfs_inode *nfsi = NFS_I(inode);
2223 struct rb_root *root_node = &nfsi->access_cache;
2224 struct rb_node **p = &root_node->rb_node;
2225 struct rb_node *parent = NULL;
2226 struct nfs_access_entry *entry;
2228 spin_lock(&inode->i_lock);
2229 while (*p != NULL) {
2231 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2233 if (set->cred < entry->cred)
2234 p = &parent->rb_left;
2235 else if (set->cred > entry->cred)
2236 p = &parent->rb_right;
2240 rb_link_node(&set->rb_node, parent, p);
2241 rb_insert_color(&set->rb_node, root_node);
2242 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2243 spin_unlock(&inode->i_lock);
2246 rb_replace_node(parent, &set->rb_node, root_node);
2247 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2248 list_del(&entry->lru);
2249 spin_unlock(&inode->i_lock);
2250 nfs_access_free_entry(entry);
2253 static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2255 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2258 RB_CLEAR_NODE(&cache->rb_node);
2259 cache->jiffies = set->jiffies;
2260 cache->cred = get_rpccred(set->cred);
2261 cache->mask = set->mask;
2263 nfs_access_add_rbtree(inode, cache);
2265 /* Update accounting */
2266 smp_mb__before_atomic_inc();
2267 atomic_long_inc(&nfs_access_nr_entries);
2268 smp_mb__after_atomic_inc();
2270 /* Add inode to global LRU list */
2271 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2272 spin_lock(&nfs_access_lru_lock);
2273 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2274 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2275 &nfs_access_lru_list);
2276 spin_unlock(&nfs_access_lru_lock);
2280 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2282 struct nfs_access_entry cache;
2285 status = nfs_access_get_cached(inode, cred, &cache);
2289 /* Be clever: ask server to check for all possible rights */
2290 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2292 cache.jiffies = jiffies;
2293 status = NFS_PROTO(inode)->access(inode, &cache);
2295 if (status == -ESTALE) {
2296 nfs_zap_caches(inode);
2297 if (!S_ISDIR(inode->i_mode))
2298 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2302 nfs_access_add_cache(inode, &cache);
2304 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2309 static int nfs_open_permission_mask(int openflags)
2313 if ((openflags & O_ACCMODE) != O_WRONLY)
2315 if ((openflags & O_ACCMODE) != O_RDONLY)
2317 if (openflags & __FMODE_EXEC)
2322 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2324 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2327 int nfs_permission(struct inode *inode, int mask)
2329 struct rpc_cred *cred;
2332 if (mask & MAY_NOT_BLOCK)
2335 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2337 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2339 /* Is this sys_access() ? */
2340 if (mask & (MAY_ACCESS | MAY_CHDIR))
2343 switch (inode->i_mode & S_IFMT) {
2347 /* NFSv4 has atomic_open... */
2348 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2349 && (mask & MAY_OPEN)
2350 && !(mask & MAY_EXEC))
2355 * Optimize away all write operations, since the server
2356 * will check permissions when we perform the op.
2358 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2363 if (!NFS_PROTO(inode)->access)
2366 cred = rpc_lookup_cred();
2367 if (!IS_ERR(cred)) {
2368 res = nfs_do_access(inode, cred, mask);
2371 res = PTR_ERR(cred);
2373 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2376 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2377 inode->i_sb->s_id, inode->i_ino, mask, res);
2380 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2382 res = generic_permission(inode, mask);
2388 * version-control: t
2389 * kept-new-versions: 5