4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
41 #include "delegation.h"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
57 const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .iterate = nfs_readdir,
62 .release = nfs_closedir,
63 .fsync = nfs_fsync_dir,
66 const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
72 struct nfs_inode *nfsi = NFS_I(dir);
73 struct nfs_open_dir_context *ctx;
74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
77 ctx->attr_gencount = nfsi->attr_gencount;
80 ctx->cred = get_rpccred(cred);
81 spin_lock(&dir->i_lock);
82 list_add(&ctx->list, &nfsi->open_files);
83 spin_unlock(&dir->i_lock);
86 return ERR_PTR(-ENOMEM);
89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
91 spin_lock(&dir->i_lock);
93 spin_unlock(&dir->i_lock);
94 put_rpccred(ctx->cred);
102 nfs_opendir(struct inode *inode, struct file *filp)
105 struct nfs_open_dir_context *ctx;
106 struct rpc_cred *cred;
108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
110 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
112 cred = rpc_lookup_cred();
114 return PTR_ERR(cred);
115 ctx = alloc_nfs_open_dir_context(inode, cred);
120 filp->private_data = ctx;
121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 /* This is a mountpoint, so d_revalidate will never
123 * have been called, so we need to refresh the
124 * inode (for close-open consistency) ourselves.
126 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
134 nfs_closedir(struct inode *inode, struct file *filp)
136 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
140 struct nfs_cache_array_entry {
144 unsigned char d_type;
147 struct nfs_cache_array {
151 struct nfs_cache_array_entry array[0];
154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
158 struct dir_context *ctx;
159 unsigned long page_index;
162 loff_t current_index;
163 decode_dirent_t decode;
165 unsigned long timestamp;
166 unsigned long gencount;
167 unsigned int cache_entry_index;
170 } nfs_readdir_descriptor_t;
173 * The caller is responsible for calling nfs_readdir_release_array(page)
176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
180 return ERR_PTR(-EIO);
183 return ERR_PTR(-ENOMEM);
188 void nfs_readdir_release_array(struct page *page)
194 * we are freeing strings created by nfs_add_to_readdir_array()
197 void nfs_readdir_clear_array(struct page *page)
199 struct nfs_cache_array *array;
202 array = kmap_atomic(page);
203 for (i = 0; i < array->size; i++)
204 kfree(array->array[i].string.name);
205 kunmap_atomic(array);
209 * the caller is responsible for freeing qstr.name
210 * when called by nfs_readdir_add_to_array, the strings will be freed in
211 * nfs_clear_readdir_array()
214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
217 string->name = kmemdup(name, len, GFP_KERNEL);
218 if (string->name == NULL)
221 * Avoid a kmemleak false positive. The pointer to the name is stored
222 * in a page cache page which kmemleak does not scan.
224 kmemleak_not_leak(string->name);
225 string->hash = full_name_hash(name, len);
230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
232 struct nfs_cache_array *array = nfs_readdir_get_array(page);
233 struct nfs_cache_array_entry *cache_entry;
237 return PTR_ERR(array);
239 cache_entry = &array->array[array->size];
241 /* Check that this entry lies within the page bounds */
243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
246 cache_entry->cookie = entry->prev_cookie;
247 cache_entry->ino = entry->ino;
248 cache_entry->d_type = entry->d_type;
249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
252 array->last_cookie = entry->cookie;
255 array->eof_index = array->size;
257 nfs_readdir_release_array(page);
262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
264 loff_t diff = desc->ctx->pos - desc->current_index;
269 if (diff >= array->size) {
270 if (array->eof_index >= 0)
275 index = (unsigned int)diff;
276 *desc->dir_cookie = array->array[index].cookie;
277 desc->cache_entry_index = index;
285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
298 int status = -EAGAIN;
300 for (i = 0; i < array->size; i++) {
301 if (array->array[i].cookie == *desc->dir_cookie) {
302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303 struct nfs_open_dir_context *ctx = desc->file->private_data;
305 new_pos = desc->current_index + i;
306 if (ctx->attr_gencount != nfsi->attr_gencount ||
307 !nfs_readdir_inode_mapping_valid(nfsi)) {
309 ctx->attr_gencount = nfsi->attr_gencount;
310 } else if (new_pos < desc->ctx->pos) {
312 && ctx->dup_cookie == *desc->dir_cookie) {
313 if (printk_ratelimit()) {
314 pr_notice("NFS: directory %pD2 contains a readdir loop."
315 "Please contact your server vendor. "
316 "The file: %.*s has duplicate cookie %llu\n",
317 desc->file, array->array[i].string.len,
318 array->array[i].string.name, *desc->dir_cookie);
323 ctx->dup_cookie = *desc->dir_cookie;
326 desc->ctx->pos = new_pos;
327 desc->cache_entry_index = i;
331 if (array->eof_index >= 0) {
332 status = -EBADCOOKIE;
333 if (*desc->dir_cookie == array->last_cookie)
341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
343 struct nfs_cache_array *array;
346 array = nfs_readdir_get_array(desc->page);
348 status = PTR_ERR(array);
352 if (*desc->dir_cookie == 0)
353 status = nfs_readdir_search_for_pos(array, desc);
355 status = nfs_readdir_search_for_cookie(array, desc);
357 if (status == -EAGAIN) {
358 desc->last_cookie = array->last_cookie;
359 desc->current_index += array->size;
362 nfs_readdir_release_array(desc->page);
367 /* Fill a page with xdr information before transferring to the cache page */
369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
370 struct nfs_entry *entry, struct file *file, struct inode *inode)
372 struct nfs_open_dir_context *ctx = file->private_data;
373 struct rpc_cred *cred = ctx->cred;
374 unsigned long timestamp, gencount;
379 gencount = nfs_inc_attr_generation_counter();
380 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
381 NFS_SERVER(inode)->dtsize, desc->plus);
383 /* We requested READDIRPLUS, but the server doesn't grok it */
384 if (error == -ENOTSUPP && desc->plus) {
385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
392 desc->timestamp = timestamp;
393 desc->gencount = gencount;
398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399 struct nfs_entry *entry, struct xdr_stream *xdr)
403 error = desc->decode(xdr, entry, desc->plus);
406 entry->fattr->time_start = desc->timestamp;
407 entry->fattr->gencount = desc->gencount;
411 /* Match file and dirent using either filehandle or fileid
412 * Note: caller is responsible for checking the fsid
415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
417 struct nfs_inode *nfsi;
419 if (d_really_is_negative(dentry))
422 nfsi = NFS_I(d_inode(dentry));
423 if (entry->fattr->fileid == nfsi->fileid)
425 if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0)
431 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
433 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
435 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
443 * This function is called by the lookup code to request the use of
444 * readdirplus to accelerate any future lookups in the same
448 void nfs_advise_use_readdirplus(struct inode *dir)
450 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
454 * This function is mainly for use by nfs_getattr().
456 * If this is an 'ls -l', we want to force use of readdirplus.
457 * Do this by checking if there is an active file descriptor
458 * and calling nfs_advise_use_readdirplus, then forcing a
461 void nfs_force_use_readdirplus(struct inode *dir)
463 if (!list_empty(&NFS_I(dir)->open_files)) {
464 nfs_advise_use_readdirplus(dir);
465 nfs_zap_mapping(dir, dir->i_mapping);
470 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
472 struct qstr filename = QSTR_INIT(entry->name, entry->len);
473 struct dentry *dentry;
474 struct dentry *alias;
475 struct inode *dir = d_inode(parent);
479 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
481 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
483 if (filename.name[0] == '.') {
484 if (filename.len == 1)
486 if (filename.len == 2 && filename.name[1] == '.')
489 filename.hash = full_name_hash(filename.name, filename.len);
491 dentry = d_lookup(parent, &filename);
492 if (dentry != NULL) {
493 /* Is there a mountpoint here? If so, just exit */
494 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
495 &entry->fattr->fsid))
497 if (nfs_same_file(dentry, entry)) {
498 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
499 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
501 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
504 d_invalidate(dentry);
509 dentry = d_alloc(parent, &filename);
513 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
517 alias = d_splice_alias(inode, dentry);
521 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
524 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
530 /* Perform conversion from xdr to cache array */
532 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
533 struct page **xdr_pages, struct page *page, unsigned int buflen)
535 struct xdr_stream stream;
537 struct page *scratch;
538 struct nfs_cache_array *array;
539 unsigned int count = 0;
542 scratch = alloc_page(GFP_KERNEL);
546 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
547 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
550 status = xdr_decode(desc, entry, &stream);
552 if (status == -EAGAIN)
560 nfs_prime_dcache(desc->file->f_path.dentry, entry);
562 status = nfs_readdir_add_to_array(entry, page);
565 } while (!entry->eof);
567 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
568 array = nfs_readdir_get_array(page);
569 if (!IS_ERR(array)) {
570 array->eof_index = array->size;
572 nfs_readdir_release_array(page);
574 status = PTR_ERR(array);
582 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
585 for (i = 0; i < npages; i++)
590 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
593 nfs_readdir_free_pagearray(pages, npages);
597 * nfs_readdir_large_page will allocate pages that must be freed with a call
598 * to nfs_readdir_free_large_page
601 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
605 for (i = 0; i < npages; i++) {
606 struct page *page = alloc_page(GFP_KERNEL);
614 nfs_readdir_free_pagearray(pages, i);
619 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
621 struct page *pages[NFS_MAX_READDIR_PAGES];
622 void *pages_ptr = NULL;
623 struct nfs_entry entry;
624 struct file *file = desc->file;
625 struct nfs_cache_array *array;
626 int status = -ENOMEM;
627 unsigned int array_size = ARRAY_SIZE(pages);
629 entry.prev_cookie = 0;
630 entry.cookie = desc->last_cookie;
632 entry.fh = nfs_alloc_fhandle();
633 entry.fattr = nfs_alloc_fattr();
634 entry.server = NFS_SERVER(inode);
635 if (entry.fh == NULL || entry.fattr == NULL)
638 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
639 if (IS_ERR(entry.label)) {
640 status = PTR_ERR(entry.label);
644 array = nfs_readdir_get_array(page);
646 status = PTR_ERR(array);
649 memset(array, 0, sizeof(struct nfs_cache_array));
650 array->eof_index = -1;
652 status = nfs_readdir_large_page(pages, array_size);
654 goto out_release_array;
657 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
662 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
664 if (status == -ENOSPC)
668 } while (array->eof_index < 0);
670 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
672 nfs_readdir_release_array(page);
674 nfs4_label_free(entry.label);
676 nfs_free_fattr(entry.fattr);
677 nfs_free_fhandle(entry.fh);
682 * Now we cache directories properly, by converting xdr information
683 * to an array that can be used for lookups later. This results in
684 * fewer cache pages, since we can store more information on each page.
685 * We only need to convert from xdr once so future lookups are much simpler
688 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
690 struct inode *inode = file_inode(desc->file);
693 ret = nfs_readdir_xdr_to_array(desc, page, inode);
696 SetPageUptodate(page);
698 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
699 /* Should never happen */
700 nfs_zap_mapping(inode, inode->i_mapping);
710 void cache_page_release(nfs_readdir_descriptor_t *desc)
712 if (!desc->page->mapping)
713 nfs_readdir_clear_array(desc->page);
714 page_cache_release(desc->page);
719 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
721 return read_cache_page(file_inode(desc->file)->i_mapping,
722 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
726 * Returns 0 if desc->dir_cookie was found on page desc->page_index
729 int find_cache_page(nfs_readdir_descriptor_t *desc)
733 desc->page = get_cache_page(desc);
734 if (IS_ERR(desc->page))
735 return PTR_ERR(desc->page);
737 res = nfs_readdir_search_array(desc);
739 cache_page_release(desc);
743 /* Search for desc->dir_cookie from the beginning of the page cache */
745 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
749 if (desc->page_index == 0) {
750 desc->current_index = 0;
751 desc->last_cookie = 0;
754 res = find_cache_page(desc);
755 } while (res == -EAGAIN);
760 * Once we've found the start of the dirent within a page: fill 'er up...
763 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
765 struct file *file = desc->file;
768 struct nfs_cache_array *array = NULL;
769 struct nfs_open_dir_context *ctx = file->private_data;
771 array = nfs_readdir_get_array(desc->page);
773 res = PTR_ERR(array);
777 for (i = desc->cache_entry_index; i < array->size; i++) {
778 struct nfs_cache_array_entry *ent;
780 ent = &array->array[i];
781 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
782 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
787 if (i < (array->size-1))
788 *desc->dir_cookie = array->array[i+1].cookie;
790 *desc->dir_cookie = array->last_cookie;
794 if (array->eof_index >= 0)
797 nfs_readdir_release_array(desc->page);
799 cache_page_release(desc);
800 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
801 (unsigned long long)*desc->dir_cookie, res);
806 * If we cannot find a cookie in our cache, we suspect that this is
807 * because it points to a deleted file, so we ask the server to return
808 * whatever it thinks is the next entry. We then feed this to filldir.
809 * If all goes well, we should then be able to find our way round the
810 * cache on the next call to readdir_search_pagecache();
812 * NOTE: we cannot add the anonymous page to the pagecache because
813 * the data it contains might not be page aligned. Besides,
814 * we should already have a complete representation of the
815 * directory in the page cache by the time we get here.
818 int uncached_readdir(nfs_readdir_descriptor_t *desc)
820 struct page *page = NULL;
822 struct inode *inode = file_inode(desc->file);
823 struct nfs_open_dir_context *ctx = desc->file->private_data;
825 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
826 (unsigned long long)*desc->dir_cookie);
828 page = alloc_page(GFP_HIGHUSER);
834 desc->page_index = 0;
835 desc->last_cookie = *desc->dir_cookie;
839 status = nfs_readdir_xdr_to_array(desc, page, inode);
843 status = nfs_do_filldir(desc);
846 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
850 cache_page_release(desc);
854 static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
856 struct nfs_inode *nfsi = NFS_I(dir);
858 if (nfs_attribute_cache_expired(dir))
860 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
865 /* The file offset position represents the dirent entry number. A
866 last cookie cache takes care of the common case of reading the
869 static int nfs_readdir(struct file *file, struct dir_context *ctx)
871 struct dentry *dentry = file->f_path.dentry;
872 struct inode *inode = d_inode(dentry);
873 nfs_readdir_descriptor_t my_desc,
875 struct nfs_open_dir_context *dir_ctx = file->private_data;
878 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
879 file, (long long)ctx->pos);
880 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
883 * ctx->pos points to the dirent entry number.
884 * *desc->dir_cookie has the cookie for the next entry. We have
885 * to either find the entry with the appropriate number or
886 * revalidate the cookie.
888 memset(desc, 0, sizeof(*desc));
892 desc->dir_cookie = &dir_ctx->dir_cookie;
893 desc->decode = NFS_PROTO(inode)->decode_dirent;
894 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
896 nfs_block_sillyrename(dentry);
897 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
898 res = nfs_revalidate_mapping(inode, file->f_mapping);
903 res = readdir_search_pagecache(desc);
905 if (res == -EBADCOOKIE) {
907 /* This means either end of directory */
908 if (*desc->dir_cookie && desc->eof == 0) {
909 /* Or that the server has 'lost' a cookie */
910 res = uncached_readdir(desc);
916 if (res == -ETOOSMALL && desc->plus) {
917 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
918 nfs_zap_caches(inode);
919 desc->page_index = 0;
927 res = nfs_do_filldir(desc);
930 } while (!desc->eof);
932 nfs_unblock_sillyrename(dentry);
935 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
939 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
941 struct inode *inode = file_inode(filp);
942 struct nfs_open_dir_context *dir_ctx = filp->private_data;
944 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
945 filp, offset, whence);
947 mutex_lock(&inode->i_mutex);
950 offset += filp->f_pos;
958 if (offset != filp->f_pos) {
959 filp->f_pos = offset;
960 dir_ctx->dir_cookie = 0;
964 mutex_unlock(&inode->i_mutex);
969 * All directory operations under NFS are synchronous, so fsync()
970 * is a dummy operation.
972 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
975 struct inode *inode = file_inode(filp);
977 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
979 mutex_lock(&inode->i_mutex);
980 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
981 mutex_unlock(&inode->i_mutex);
986 * nfs_force_lookup_revalidate - Mark the directory as having changed
987 * @dir - pointer to directory inode
989 * This forces the revalidation code in nfs_lookup_revalidate() to do a
990 * full lookup on all child dentries of 'dir' whenever a change occurs
991 * on the server that might have invalidated our dcache.
993 * The caller should be holding dir->i_lock
995 void nfs_force_lookup_revalidate(struct inode *dir)
997 NFS_I(dir)->cache_change_attribute++;
999 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1002 * A check for whether or not the parent directory has changed.
1003 * In the case it has, we assume that the dentries are untrustworthy
1004 * and may need to be looked up again.
1005 * If rcu_walk prevents us from performing a full check, return 0.
1007 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1012 if (IS_ROOT(dentry))
1014 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1016 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1018 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1020 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir);
1022 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1025 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1031 * Use intent information to check whether or not we're going to do
1032 * an O_EXCL create using this path component.
1034 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1036 if (NFS_PROTO(dir)->version == 2)
1038 return flags & LOOKUP_EXCL;
1042 * Inode and filehandle revalidation for lookups.
1044 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1045 * or if the intent information indicates that we're about to open this
1046 * particular file and the "nocto" mount flag is not set.
1050 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1052 struct nfs_server *server = NFS_SERVER(inode);
1055 if (IS_AUTOMOUNT(inode))
1057 /* VFS wants an on-the-wire revalidation */
1058 if (flags & LOOKUP_REVAL)
1060 /* This is an open(2) */
1061 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1062 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1065 return (inode->i_nlink == 0) ? -ENOENT : 0;
1067 if (flags & LOOKUP_RCU)
1069 ret = __nfs_revalidate_inode(server, inode);
1076 * We judge how long we want to trust negative
1077 * dentries by looking at the parent inode mtime.
1079 * If parent mtime has changed, we revalidate, else we wait for a
1080 * period corresponding to the parent's attribute cache timeout value.
1082 * If LOOKUP_RCU prevents us from performing a full check, return 1
1083 * suggesting a reval is needed.
1086 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1089 /* Don't revalidate a negative dentry if we're creating a new file */
1090 if (flags & LOOKUP_CREATE)
1092 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1094 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1098 * This is called every time the dcache has a lookup hit,
1099 * and we should check whether we can really trust that
1102 * NOTE! The hit can be a negative hit too, don't assume
1105 * If the parent directory is seen to have changed, we throw out the
1106 * cached dentry and do a new lookup.
1108 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1111 struct inode *inode;
1112 struct dentry *parent;
1113 struct nfs_fh *fhandle = NULL;
1114 struct nfs_fattr *fattr = NULL;
1115 struct nfs4_label *label = NULL;
1118 if (flags & LOOKUP_RCU) {
1119 parent = ACCESS_ONCE(dentry->d_parent);
1120 dir = d_inode_rcu(parent);
1124 parent = dget_parent(dentry);
1125 dir = d_inode(parent);
1127 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1128 inode = d_inode(dentry);
1131 if (nfs_neg_need_reval(dir, dentry, flags)) {
1132 if (flags & LOOKUP_RCU)
1136 goto out_valid_noent;
1139 if (is_bad_inode(inode)) {
1140 if (flags & LOOKUP_RCU)
1142 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1147 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1148 goto out_set_verifier;
1150 /* Force a full look up iff the parent directory has changed */
1151 if (!nfs_is_exclusive_create(dir, flags) &&
1152 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1154 if (nfs_lookup_verify_inode(inode, flags)) {
1155 if (flags & LOOKUP_RCU)
1157 goto out_zap_parent;
1162 if (flags & LOOKUP_RCU)
1165 if (NFS_STALE(inode))
1169 fhandle = nfs_alloc_fhandle();
1170 fattr = nfs_alloc_fattr();
1171 if (fhandle == NULL || fattr == NULL)
1174 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1178 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1179 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1180 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1183 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1185 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1188 nfs_setsecurity(inode, fattr, label);
1190 nfs_free_fattr(fattr);
1191 nfs_free_fhandle(fhandle);
1192 nfs4_label_free(label);
1195 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1197 /* Success: notify readdir to use READDIRPLUS */
1198 nfs_advise_use_readdirplus(dir);
1200 if (flags & LOOKUP_RCU) {
1201 if (parent != ACCESS_ONCE(dentry->d_parent))
1205 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1209 nfs_zap_caches(dir);
1211 WARN_ON(flags & LOOKUP_RCU);
1212 nfs_free_fattr(fattr);
1213 nfs_free_fhandle(fhandle);
1214 nfs4_label_free(label);
1215 nfs_mark_for_revalidate(dir);
1216 if (inode && S_ISDIR(inode->i_mode)) {
1217 /* Purge readdir caches. */
1218 nfs_zap_caches(inode);
1220 * We can't d_drop the root of a disconnected tree:
1221 * its d_hash is on the s_anon list and d_drop() would hide
1222 * it from shrink_dcache_for_unmount(), leading to busy
1223 * inodes on unmount and further oopses.
1225 if (IS_ROOT(dentry))
1229 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1233 WARN_ON(flags & LOOKUP_RCU);
1234 nfs_free_fattr(fattr);
1235 nfs_free_fhandle(fhandle);
1236 nfs4_label_free(label);
1238 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1239 __func__, dentry, error);
1244 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1245 * when we don't really care about the dentry name. This is called when a
1246 * pathwalk ends on a dentry that was not found via a normal lookup in the
1247 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1249 * In this situation, we just want to verify that the inode itself is OK
1250 * since the dentry might have changed on the server.
1252 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1255 struct inode *inode = d_inode(dentry);
1258 * I believe we can only get a negative dentry here in the case of a
1259 * procfs-style symlink. Just assume it's correct for now, but we may
1260 * eventually need to do something more here.
1263 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1268 if (is_bad_inode(inode)) {
1269 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1274 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1275 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1276 __func__, inode->i_ino, error ? "invalid" : "valid");
1281 * This is called from dput() when d_count is going to 0.
1283 static int nfs_dentry_delete(const struct dentry *dentry)
1285 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1286 dentry, dentry->d_flags);
1288 /* Unhash any dentry with a stale inode */
1289 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1292 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1293 /* Unhash it, so that ->d_iput() would be called */
1296 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1297 /* Unhash it, so that ancestors of killed async unlink
1298 * files will be cleaned up during umount */
1305 /* Ensure that we revalidate inode->i_nlink */
1306 static void nfs_drop_nlink(struct inode *inode)
1308 spin_lock(&inode->i_lock);
1309 /* drop the inode if we're reasonably sure this is the last link */
1310 if (inode->i_nlink == 1)
1312 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1313 spin_unlock(&inode->i_lock);
1317 * Called when the dentry loses inode.
1318 * We use it to clean up silly-renamed files.
1320 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1322 if (S_ISDIR(inode->i_mode))
1323 /* drop any readdir cache as it could easily be old */
1324 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1326 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1327 nfs_complete_unlink(dentry, inode);
1328 nfs_drop_nlink(inode);
1333 static void nfs_d_release(struct dentry *dentry)
1335 /* free cached devname value, if it survived that far */
1336 if (unlikely(dentry->d_fsdata)) {
1337 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1340 kfree(dentry->d_fsdata);
1344 const struct dentry_operations nfs_dentry_operations = {
1345 .d_revalidate = nfs_lookup_revalidate,
1346 .d_weak_revalidate = nfs_weak_revalidate,
1347 .d_delete = nfs_dentry_delete,
1348 .d_iput = nfs_dentry_iput,
1349 .d_automount = nfs_d_automount,
1350 .d_release = nfs_d_release,
1352 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1354 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1357 struct dentry *parent;
1358 struct inode *inode = NULL;
1359 struct nfs_fh *fhandle = NULL;
1360 struct nfs_fattr *fattr = NULL;
1361 struct nfs4_label *label = NULL;
1364 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1365 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1367 res = ERR_PTR(-ENAMETOOLONG);
1368 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1372 * If we're doing an exclusive create, optimize away the lookup
1373 * but don't hash the dentry.
1375 if (nfs_is_exclusive_create(dir, flags)) {
1376 d_instantiate(dentry, NULL);
1381 res = ERR_PTR(-ENOMEM);
1382 fhandle = nfs_alloc_fhandle();
1383 fattr = nfs_alloc_fattr();
1384 if (fhandle == NULL || fattr == NULL)
1387 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1391 parent = dentry->d_parent;
1392 /* Protect against concurrent sillydeletes */
1393 trace_nfs_lookup_enter(dir, dentry, flags);
1394 nfs_block_sillyrename(parent);
1395 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1396 if (error == -ENOENT)
1399 res = ERR_PTR(error);
1400 goto out_unblock_sillyrename;
1402 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1403 res = ERR_CAST(inode);
1405 goto out_unblock_sillyrename;
1407 /* Success: notify readdir to use READDIRPLUS */
1408 nfs_advise_use_readdirplus(dir);
1411 res = d_splice_alias(inode, dentry);
1414 goto out_unblock_sillyrename;
1417 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1418 out_unblock_sillyrename:
1419 nfs_unblock_sillyrename(parent);
1420 trace_nfs_lookup_exit(dir, dentry, flags, error);
1421 nfs4_label_free(label);
1423 nfs_free_fattr(fattr);
1424 nfs_free_fhandle(fhandle);
1427 EXPORT_SYMBOL_GPL(nfs_lookup);
1429 #if IS_ENABLED(CONFIG_NFS_V4)
1430 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1432 const struct dentry_operations nfs4_dentry_operations = {
1433 .d_revalidate = nfs4_lookup_revalidate,
1434 .d_delete = nfs_dentry_delete,
1435 .d_iput = nfs_dentry_iput,
1436 .d_automount = nfs_d_automount,
1437 .d_release = nfs_d_release,
1439 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1441 static fmode_t flags_to_mode(int flags)
1443 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1444 if ((flags & O_ACCMODE) != O_WRONLY)
1446 if ((flags & O_ACCMODE) != O_RDONLY)
1451 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1453 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1456 static int do_open(struct inode *inode, struct file *filp)
1458 nfs_fscache_open_file(inode, filp);
1462 static int nfs_finish_open(struct nfs_open_context *ctx,
1463 struct dentry *dentry,
1464 struct file *file, unsigned open_flags,
1469 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1470 *opened |= FILE_CREATED;
1472 err = finish_open(file, dentry, do_open, opened);
1475 nfs_file_set_open_context(file, ctx);
1481 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1482 struct file *file, unsigned open_flags,
1483 umode_t mode, int *opened)
1485 struct nfs_open_context *ctx;
1487 struct iattr attr = { .ia_valid = ATTR_OPEN };
1488 struct inode *inode;
1489 unsigned int lookup_flags = 0;
1492 /* Expect a negative dentry */
1493 BUG_ON(d_inode(dentry));
1495 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1496 dir->i_sb->s_id, dir->i_ino, dentry);
1498 err = nfs_check_flags(open_flags);
1502 /* NFS only supports OPEN on regular files */
1503 if ((open_flags & O_DIRECTORY)) {
1504 if (!d_unhashed(dentry)) {
1506 * Hashed negative dentry with O_DIRECTORY: dentry was
1507 * revalidated and is fine, no need to perform lookup
1512 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1516 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1517 return -ENAMETOOLONG;
1519 if (open_flags & O_CREAT) {
1520 attr.ia_valid |= ATTR_MODE;
1521 attr.ia_mode = mode & ~current_umask();
1523 if (open_flags & O_TRUNC) {
1524 attr.ia_valid |= ATTR_SIZE;
1528 ctx = create_nfs_open_context(dentry, open_flags);
1533 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1534 nfs_block_sillyrename(dentry->d_parent);
1535 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1536 nfs_unblock_sillyrename(dentry->d_parent);
1537 if (IS_ERR(inode)) {
1538 err = PTR_ERR(inode);
1539 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1540 put_nfs_open_context(ctx);
1544 d_add(dentry, NULL);
1545 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1551 if (!(open_flags & O_NOFOLLOW))
1561 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1562 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1563 put_nfs_open_context(ctx);
1568 res = nfs_lookup(dir, dentry, lookup_flags);
1573 return finish_no_open(file, res);
1575 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1577 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1579 struct inode *inode;
1582 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1584 if (d_mountpoint(dentry))
1586 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1589 inode = d_inode(dentry);
1591 /* We can't create new files in nfs_open_revalidate(), so we
1592 * optimize away revalidation of negative dentries.
1594 if (inode == NULL) {
1595 struct dentry *parent;
1598 if (flags & LOOKUP_RCU) {
1599 parent = ACCESS_ONCE(dentry->d_parent);
1600 dir = d_inode_rcu(parent);
1604 parent = dget_parent(dentry);
1605 dir = d_inode(parent);
1607 if (!nfs_neg_need_reval(dir, dentry, flags))
1609 else if (flags & LOOKUP_RCU)
1611 if (!(flags & LOOKUP_RCU))
1613 else if (parent != ACCESS_ONCE(dentry->d_parent))
1618 /* NFS only supports OPEN on regular files */
1619 if (!S_ISREG(inode->i_mode))
1621 /* We cannot do exclusive creation on a positive dentry */
1622 if (flags & LOOKUP_EXCL)
1625 /* Let f_op->open() actually open (and revalidate) the file */
1632 return nfs_lookup_revalidate(dentry, flags);
1635 #endif /* CONFIG_NFSV4 */
1638 * Code common to create, mkdir, and mknod.
1640 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1641 struct nfs_fattr *fattr,
1642 struct nfs4_label *label)
1644 struct dentry *parent = dget_parent(dentry);
1645 struct inode *dir = d_inode(parent);
1646 struct inode *inode;
1647 int error = -EACCES;
1651 /* We may have been initialized further down */
1652 if (d_really_is_positive(dentry))
1654 if (fhandle->size == 0) {
1655 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1659 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1660 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1661 struct nfs_server *server = NFS_SB(dentry->d_sb);
1662 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1666 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1667 error = PTR_ERR(inode);
1670 d_add(dentry, inode);
1675 nfs_mark_for_revalidate(dir);
1679 EXPORT_SYMBOL_GPL(nfs_instantiate);
1682 * Following a failed create operation, we drop the dentry rather
1683 * than retain a negative dentry. This avoids a problem in the event
1684 * that the operation succeeded on the server, but an error in the
1685 * reply path made it appear to have failed.
1687 int nfs_create(struct inode *dir, struct dentry *dentry,
1688 umode_t mode, bool excl)
1691 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1694 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1695 dir->i_sb->s_id, dir->i_ino, dentry);
1697 attr.ia_mode = mode;
1698 attr.ia_valid = ATTR_MODE;
1700 trace_nfs_create_enter(dir, dentry, open_flags);
1701 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1702 trace_nfs_create_exit(dir, dentry, open_flags, error);
1710 EXPORT_SYMBOL_GPL(nfs_create);
1713 * See comments for nfs_proc_create regarding failed operations.
1716 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1721 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1722 dir->i_sb->s_id, dir->i_ino, dentry);
1724 if (!new_valid_dev(rdev))
1727 attr.ia_mode = mode;
1728 attr.ia_valid = ATTR_MODE;
1730 trace_nfs_mknod_enter(dir, dentry);
1731 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1732 trace_nfs_mknod_exit(dir, dentry, status);
1740 EXPORT_SYMBOL_GPL(nfs_mknod);
1743 * See comments for nfs_proc_create regarding failed operations.
1745 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1750 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1751 dir->i_sb->s_id, dir->i_ino, dentry);
1753 attr.ia_valid = ATTR_MODE;
1754 attr.ia_mode = mode | S_IFDIR;
1756 trace_nfs_mkdir_enter(dir, dentry);
1757 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1758 trace_nfs_mkdir_exit(dir, dentry, error);
1766 EXPORT_SYMBOL_GPL(nfs_mkdir);
1768 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1770 if (d_really_is_positive(dentry) && !d_unhashed(dentry))
1774 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1778 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1779 dir->i_sb->s_id, dir->i_ino, dentry);
1781 trace_nfs_rmdir_enter(dir, dentry);
1782 if (d_really_is_positive(dentry)) {
1783 nfs_wait_on_sillyrename(dentry);
1784 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1785 /* Ensure the VFS deletes this inode */
1788 clear_nlink(d_inode(dentry));
1791 nfs_dentry_handle_enoent(dentry);
1794 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1795 trace_nfs_rmdir_exit(dir, dentry, error);
1799 EXPORT_SYMBOL_GPL(nfs_rmdir);
1802 * Remove a file after making sure there are no pending writes,
1803 * and after checking that the file has only one user.
1805 * We invalidate the attribute cache and free the inode prior to the operation
1806 * to avoid possible races if the server reuses the inode.
1808 static int nfs_safe_remove(struct dentry *dentry)
1810 struct inode *dir = d_inode(dentry->d_parent);
1811 struct inode *inode = d_inode(dentry);
1814 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1816 /* If the dentry was sillyrenamed, we simply call d_delete() */
1817 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1822 trace_nfs_remove_enter(dir, dentry);
1823 if (inode != NULL) {
1824 NFS_PROTO(inode)->return_delegation(inode);
1825 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1827 nfs_drop_nlink(inode);
1829 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1830 if (error == -ENOENT)
1831 nfs_dentry_handle_enoent(dentry);
1832 trace_nfs_remove_exit(dir, dentry, error);
1837 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1838 * belongs to an active ".nfs..." file and we return -EBUSY.
1840 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1842 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1845 int need_rehash = 0;
1847 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1848 dir->i_ino, dentry);
1850 trace_nfs_unlink_enter(dir, dentry);
1851 spin_lock(&dentry->d_lock);
1852 if (d_count(dentry) > 1) {
1853 spin_unlock(&dentry->d_lock);
1854 /* Start asynchronous writeout of the inode */
1855 write_inode_now(d_inode(dentry), 0);
1856 error = nfs_sillyrename(dir, dentry);
1859 if (!d_unhashed(dentry)) {
1863 spin_unlock(&dentry->d_lock);
1864 error = nfs_safe_remove(dentry);
1865 if (!error || error == -ENOENT) {
1866 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1867 } else if (need_rehash)
1870 trace_nfs_unlink_exit(dir, dentry, error);
1873 EXPORT_SYMBOL_GPL(nfs_unlink);
1876 * To create a symbolic link, most file systems instantiate a new inode,
1877 * add a page to it containing the path, then write it out to the disk
1878 * using prepare_write/commit_write.
1880 * Unfortunately the NFS client can't create the in-core inode first
1881 * because it needs a file handle to create an in-core inode (see
1882 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1883 * symlink request has completed on the server.
1885 * So instead we allocate a raw page, copy the symname into it, then do
1886 * the SYMLINK request with the page as the buffer. If it succeeds, we
1887 * now have a new file handle and can instantiate an in-core NFS inode
1888 * and move the raw page into its mapping.
1890 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1895 unsigned int pathlen = strlen(symname);
1898 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1899 dir->i_ino, dentry, symname);
1901 if (pathlen > PAGE_SIZE)
1902 return -ENAMETOOLONG;
1904 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1905 attr.ia_valid = ATTR_MODE;
1907 page = alloc_page(GFP_HIGHUSER);
1911 kaddr = kmap_atomic(page);
1912 memcpy(kaddr, symname, pathlen);
1913 if (pathlen < PAGE_SIZE)
1914 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1915 kunmap_atomic(kaddr);
1917 trace_nfs_symlink_enter(dir, dentry);
1918 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1919 trace_nfs_symlink_exit(dir, dentry, error);
1921 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1922 dir->i_sb->s_id, dir->i_ino,
1923 dentry, symname, error);
1930 * No big deal if we can't add this page to the page cache here.
1931 * READLINK will get the missing page from the server if needed.
1933 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1935 SetPageUptodate(page);
1938 * add_to_page_cache_lru() grabs an extra page refcount.
1939 * Drop it here to avoid leaking this page later.
1941 page_cache_release(page);
1947 EXPORT_SYMBOL_GPL(nfs_symlink);
1950 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1952 struct inode *inode = d_inode(old_dentry);
1955 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1956 old_dentry, dentry);
1958 trace_nfs_link_enter(inode, dir, dentry);
1959 NFS_PROTO(inode)->return_delegation(inode);
1962 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1965 d_add(dentry, inode);
1967 trace_nfs_link_exit(inode, dir, dentry, error);
1970 EXPORT_SYMBOL_GPL(nfs_link);
1974 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1975 * different file handle for the same inode after a rename (e.g. when
1976 * moving to a different directory). A fail-safe method to do so would
1977 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1978 * rename the old file using the sillyrename stuff. This way, the original
1979 * file in old_dir will go away when the last process iput()s the inode.
1983 * It actually works quite well. One needs to have the possibility for
1984 * at least one ".nfs..." file in each directory the file ever gets
1985 * moved or linked to which happens automagically with the new
1986 * implementation that only depends on the dcache stuff instead of
1987 * using the inode layer
1989 * Unfortunately, things are a little more complicated than indicated
1990 * above. For a cross-directory move, we want to make sure we can get
1991 * rid of the old inode after the operation. This means there must be
1992 * no pending writes (if it's a file), and the use count must be 1.
1993 * If these conditions are met, we can drop the dentries before doing
1996 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1997 struct inode *new_dir, struct dentry *new_dentry)
1999 struct inode *old_inode = d_inode(old_dentry);
2000 struct inode *new_inode = d_inode(new_dentry);
2001 struct dentry *dentry = NULL, *rehash = NULL;
2002 struct rpc_task *task;
2005 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2006 old_dentry, new_dentry,
2007 d_count(new_dentry));
2009 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2011 * For non-directories, check whether the target is busy and if so,
2012 * make a copy of the dentry and then do a silly-rename. If the
2013 * silly-rename succeeds, the copied dentry is hashed and becomes
2016 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2018 * To prevent any new references to the target during the
2019 * rename, we unhash the dentry in advance.
2021 if (!d_unhashed(new_dentry)) {
2023 rehash = new_dentry;
2026 if (d_count(new_dentry) > 2) {
2029 /* copy the target dentry's name */
2030 dentry = d_alloc(new_dentry->d_parent,
2031 &new_dentry->d_name);
2035 /* silly-rename the existing target ... */
2036 err = nfs_sillyrename(new_dir, new_dentry);
2040 new_dentry = dentry;
2046 NFS_PROTO(old_inode)->return_delegation(old_inode);
2047 if (new_inode != NULL)
2048 NFS_PROTO(new_inode)->return_delegation(new_inode);
2050 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2052 error = PTR_ERR(task);
2056 error = rpc_wait_for_completion_task(task);
2058 error = task->tk_status;
2060 nfs_mark_for_revalidate(old_inode);
2064 trace_nfs_rename_exit(old_dir, old_dentry,
2065 new_dir, new_dentry, error);
2067 if (new_inode != NULL)
2068 nfs_drop_nlink(new_inode);
2069 d_move(old_dentry, new_dentry);
2070 nfs_set_verifier(new_dentry,
2071 nfs_save_change_attribute(new_dir));
2072 } else if (error == -ENOENT)
2073 nfs_dentry_handle_enoent(old_dentry);
2075 /* new dentry created? */
2080 EXPORT_SYMBOL_GPL(nfs_rename);
2082 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2083 static LIST_HEAD(nfs_access_lru_list);
2084 static atomic_long_t nfs_access_nr_entries;
2086 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2087 module_param(nfs_access_max_cachesize, ulong, 0644);
2088 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2090 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2092 put_rpccred(entry->cred);
2093 kfree_rcu(entry, rcu_head);
2094 smp_mb__before_atomic();
2095 atomic_long_dec(&nfs_access_nr_entries);
2096 smp_mb__after_atomic();
2099 static void nfs_access_free_list(struct list_head *head)
2101 struct nfs_access_entry *cache;
2103 while (!list_empty(head)) {
2104 cache = list_entry(head->next, struct nfs_access_entry, lru);
2105 list_del(&cache->lru);
2106 nfs_access_free_entry(cache);
2110 static unsigned long
2111 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2114 struct nfs_inode *nfsi, *next;
2115 struct nfs_access_entry *cache;
2118 spin_lock(&nfs_access_lru_lock);
2119 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2120 struct inode *inode;
2122 if (nr_to_scan-- == 0)
2124 inode = &nfsi->vfs_inode;
2125 spin_lock(&inode->i_lock);
2126 if (list_empty(&nfsi->access_cache_entry_lru))
2127 goto remove_lru_entry;
2128 cache = list_entry(nfsi->access_cache_entry_lru.next,
2129 struct nfs_access_entry, lru);
2130 list_move(&cache->lru, &head);
2131 rb_erase(&cache->rb_node, &nfsi->access_cache);
2133 if (!list_empty(&nfsi->access_cache_entry_lru))
2134 list_move_tail(&nfsi->access_cache_inode_lru,
2135 &nfs_access_lru_list);
2138 list_del_init(&nfsi->access_cache_inode_lru);
2139 smp_mb__before_atomic();
2140 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2141 smp_mb__after_atomic();
2143 spin_unlock(&inode->i_lock);
2145 spin_unlock(&nfs_access_lru_lock);
2146 nfs_access_free_list(&head);
2151 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2153 int nr_to_scan = sc->nr_to_scan;
2154 gfp_t gfp_mask = sc->gfp_mask;
2156 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2158 return nfs_do_access_cache_scan(nr_to_scan);
2163 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2165 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2169 nfs_access_cache_enforce_limit(void)
2171 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2173 unsigned int nr_to_scan;
2175 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2178 diff = nr_entries - nfs_access_max_cachesize;
2179 if (diff < nr_to_scan)
2181 nfs_do_access_cache_scan(nr_to_scan);
2184 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2186 struct rb_root *root_node = &nfsi->access_cache;
2188 struct nfs_access_entry *entry;
2190 /* Unhook entries from the cache */
2191 while ((n = rb_first(root_node)) != NULL) {
2192 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2193 rb_erase(n, root_node);
2194 list_move(&entry->lru, head);
2196 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2199 void nfs_access_zap_cache(struct inode *inode)
2203 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2205 /* Remove from global LRU init */
2206 spin_lock(&nfs_access_lru_lock);
2207 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2208 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2210 spin_lock(&inode->i_lock);
2211 __nfs_access_zap_cache(NFS_I(inode), &head);
2212 spin_unlock(&inode->i_lock);
2213 spin_unlock(&nfs_access_lru_lock);
2214 nfs_access_free_list(&head);
2216 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2218 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2220 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2221 struct nfs_access_entry *entry;
2224 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2226 if (cred < entry->cred)
2228 else if (cred > entry->cred)
2236 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2238 struct nfs_inode *nfsi = NFS_I(inode);
2239 struct nfs_access_entry *cache;
2242 spin_lock(&inode->i_lock);
2243 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2245 cache = nfs_access_search_rbtree(inode, cred);
2248 if (!nfs_have_delegated_attributes(inode) &&
2249 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2251 res->jiffies = cache->jiffies;
2252 res->cred = cache->cred;
2253 res->mask = cache->mask;
2254 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2257 spin_unlock(&inode->i_lock);
2260 rb_erase(&cache->rb_node, &nfsi->access_cache);
2261 list_del(&cache->lru);
2262 spin_unlock(&inode->i_lock);
2263 nfs_access_free_entry(cache);
2266 spin_unlock(&inode->i_lock);
2267 nfs_access_zap_cache(inode);
2271 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2273 /* Only check the most recently returned cache entry,
2274 * but do it without locking.
2276 struct nfs_inode *nfsi = NFS_I(inode);
2277 struct nfs_access_entry *cache;
2279 struct list_head *lh;
2282 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2284 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2285 cache = list_entry(lh, struct nfs_access_entry, lru);
2286 if (lh == &nfsi->access_cache_entry_lru ||
2287 cred != cache->cred)
2291 if (!nfs_have_delegated_attributes(inode) &&
2292 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2294 res->jiffies = cache->jiffies;
2295 res->cred = cache->cred;
2296 res->mask = cache->mask;
2303 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2305 struct nfs_inode *nfsi = NFS_I(inode);
2306 struct rb_root *root_node = &nfsi->access_cache;
2307 struct rb_node **p = &root_node->rb_node;
2308 struct rb_node *parent = NULL;
2309 struct nfs_access_entry *entry;
2311 spin_lock(&inode->i_lock);
2312 while (*p != NULL) {
2314 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2316 if (set->cred < entry->cred)
2317 p = &parent->rb_left;
2318 else if (set->cred > entry->cred)
2319 p = &parent->rb_right;
2323 rb_link_node(&set->rb_node, parent, p);
2324 rb_insert_color(&set->rb_node, root_node);
2325 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2326 spin_unlock(&inode->i_lock);
2329 rb_replace_node(parent, &set->rb_node, root_node);
2330 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2331 list_del(&entry->lru);
2332 spin_unlock(&inode->i_lock);
2333 nfs_access_free_entry(entry);
2336 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2338 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2341 RB_CLEAR_NODE(&cache->rb_node);
2342 cache->jiffies = set->jiffies;
2343 cache->cred = get_rpccred(set->cred);
2344 cache->mask = set->mask;
2346 /* The above field assignments must be visible
2347 * before this item appears on the lru. We cannot easily
2348 * use rcu_assign_pointer, so just force the memory barrier.
2351 nfs_access_add_rbtree(inode, cache);
2353 /* Update accounting */
2354 smp_mb__before_atomic();
2355 atomic_long_inc(&nfs_access_nr_entries);
2356 smp_mb__after_atomic();
2358 /* Add inode to global LRU list */
2359 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2360 spin_lock(&nfs_access_lru_lock);
2361 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2362 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2363 &nfs_access_lru_list);
2364 spin_unlock(&nfs_access_lru_lock);
2366 nfs_access_cache_enforce_limit();
2368 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2370 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2373 if (access_result & NFS4_ACCESS_READ)
2374 entry->mask |= MAY_READ;
2376 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2377 entry->mask |= MAY_WRITE;
2378 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2379 entry->mask |= MAY_EXEC;
2381 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2383 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2385 struct nfs_access_entry cache;
2388 trace_nfs_access_enter(inode);
2390 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2392 status = nfs_access_get_cached(inode, cred, &cache);
2397 if (mask & MAY_NOT_BLOCK)
2400 /* Be clever: ask server to check for all possible rights */
2401 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2403 cache.jiffies = jiffies;
2404 status = NFS_PROTO(inode)->access(inode, &cache);
2406 if (status == -ESTALE) {
2407 nfs_zap_caches(inode);
2408 if (!S_ISDIR(inode->i_mode))
2409 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2413 nfs_access_add_cache(inode, &cache);
2415 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2418 trace_nfs_access_exit(inode, status);
2422 static int nfs_open_permission_mask(int openflags)
2426 if (openflags & __FMODE_EXEC) {
2427 /* ONLY check exec rights */
2430 if ((openflags & O_ACCMODE) != O_WRONLY)
2432 if ((openflags & O_ACCMODE) != O_RDONLY)
2439 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2441 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2443 EXPORT_SYMBOL_GPL(nfs_may_open);
2445 int nfs_permission(struct inode *inode, int mask)
2447 struct rpc_cred *cred;
2450 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2452 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2454 /* Is this sys_access() ? */
2455 if (mask & (MAY_ACCESS | MAY_CHDIR))
2458 switch (inode->i_mode & S_IFMT) {
2465 * Optimize away all write operations, since the server
2466 * will check permissions when we perform the op.
2468 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2473 if (!NFS_PROTO(inode)->access)
2476 /* Always try fast lookups first */
2478 cred = rpc_lookup_cred_nonblock();
2480 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2482 res = PTR_ERR(cred);
2484 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2485 /* Fast lookup failed, try the slow way */
2486 cred = rpc_lookup_cred();
2487 if (!IS_ERR(cred)) {
2488 res = nfs_do_access(inode, cred, mask);
2491 res = PTR_ERR(cred);
2494 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2497 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2498 inode->i_sb->s_id, inode->i_ino, mask, res);
2501 if (mask & MAY_NOT_BLOCK)
2504 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2506 res = generic_permission(inode, mask);
2509 EXPORT_SYMBOL_GPL(nfs_permission);
2513 * version-control: t
2514 * kept-new-versions: 5