1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs directory handling functions
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
21 #include <linux/module.h>
22 #include <linux/time.h>
23 #include <linux/errno.h>
24 #include <linux/stat.h>
25 #include <linux/fcntl.h>
26 #include <linux/string.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/nfs_fs.h>
32 #include <linux/nfs_mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/pagevec.h>
35 #include <linux/namei.h>
36 #include <linux/mount.h>
37 #include <linux/swap.h>
38 #include <linux/sched.h>
39 #include <linux/kmemleak.h>
40 #include <linux/xattr.h>
42 #include "delegation.h"
49 /* #define NFS_DEBUG_VERBOSE 1 */
51 static int nfs_opendir(struct inode *, struct file *);
52 static int nfs_closedir(struct inode *, struct file *);
53 static int nfs_readdir(struct file *, struct dir_context *);
54 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56 static void nfs_readdir_clear_array(struct page*);
58 const struct file_operations nfs_dir_operations = {
59 .llseek = nfs_llseek_dir,
60 .read = generic_read_dir,
61 .iterate = nfs_readdir,
63 .release = nfs_closedir,
64 .fsync = nfs_fsync_dir,
67 const struct address_space_operations nfs_dir_aops = {
68 .freepage = nfs_readdir_clear_array,
71 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, const struct cred *cred)
73 struct nfs_inode *nfsi = NFS_I(dir);
74 struct nfs_open_dir_context *ctx;
75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
78 ctx->attr_gencount = nfsi->attr_gencount;
81 ctx->cred = get_cred(cred);
82 spin_lock(&dir->i_lock);
83 if (list_empty(&nfsi->open_files) &&
84 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
85 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
87 list_add(&ctx->list, &nfsi->open_files);
88 spin_unlock(&dir->i_lock);
91 return ERR_PTR(-ENOMEM);
94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
96 spin_lock(&dir->i_lock);
98 spin_unlock(&dir->i_lock);
107 nfs_opendir(struct inode *inode, struct file *filp)
110 struct nfs_open_dir_context *ctx;
112 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
114 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
116 ctx = alloc_nfs_open_dir_context(inode, current_cred());
121 filp->private_data = ctx;
127 nfs_closedir(struct inode *inode, struct file *filp)
129 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
133 struct nfs_cache_array_entry {
137 unsigned char d_type;
140 struct nfs_cache_array {
144 struct nfs_cache_array_entry array[0];
150 struct page *pages[NFS_MAX_READDIR_RAPAGES];
153 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool);
157 struct dir_context *ctx;
158 unsigned long page_index;
159 struct readdirvec pvec;
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 * we are freeing strings created by nfs_add_to_readdir_array()
176 void nfs_readdir_clear_array(struct page *page)
178 struct nfs_cache_array *array;
181 array = kmap_atomic(page);
182 for (i = 0; i < array->size; i++)
183 kfree(array->array[i].string.name);
184 kunmap_atomic(array);
188 * the caller is responsible for freeing qstr.name
189 * when called by nfs_readdir_add_to_array, the strings will be freed in
190 * nfs_clear_readdir_array()
193 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
196 string->name = kmemdup(name, len, GFP_KERNEL);
197 if (string->name == NULL)
200 * Avoid a kmemleak false positive. The pointer to the name is stored
201 * in a page cache page which kmemleak does not scan.
203 kmemleak_not_leak(string->name);
204 string->hash = full_name_hash(NULL, name, len);
209 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
211 struct nfs_cache_array *array = kmap(page);
212 struct nfs_cache_array_entry *cache_entry;
215 cache_entry = &array->array[array->size];
217 /* Check that this entry lies within the page bounds */
219 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
222 cache_entry->cookie = entry->prev_cookie;
223 cache_entry->ino = entry->ino;
224 cache_entry->d_type = entry->d_type;
225 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
228 array->last_cookie = entry->cookie;
231 array->eof_index = array->size;
238 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
240 loff_t diff = desc->ctx->pos - desc->current_index;
245 if (diff >= array->size) {
246 if (array->eof_index >= 0)
251 index = (unsigned int)diff;
252 *desc->dir_cookie = array->array[index].cookie;
253 desc->cache_entry_index = index;
261 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
263 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
266 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
270 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
274 int status = -EAGAIN;
276 for (i = 0; i < array->size; i++) {
277 if (array->array[i].cookie == *desc->dir_cookie) {
278 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
279 struct nfs_open_dir_context *ctx = desc->file->private_data;
281 new_pos = desc->current_index + i;
282 if (ctx->attr_gencount != nfsi->attr_gencount ||
283 !nfs_readdir_inode_mapping_valid(nfsi)) {
285 ctx->attr_gencount = nfsi->attr_gencount;
286 } else if (new_pos < desc->ctx->pos) {
288 && ctx->dup_cookie == *desc->dir_cookie) {
289 if (printk_ratelimit()) {
290 pr_notice("NFS: directory %pD2 contains a readdir loop."
291 "Please contact your server vendor. "
292 "The file: %.*s has duplicate cookie %llu\n",
293 desc->file, array->array[i].string.len,
294 array->array[i].string.name, *desc->dir_cookie);
299 ctx->dup_cookie = *desc->dir_cookie;
302 desc->ctx->pos = new_pos;
303 desc->cache_entry_index = i;
307 if (array->eof_index >= 0) {
308 status = -EBADCOOKIE;
309 if (*desc->dir_cookie == array->last_cookie)
317 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
319 struct nfs_cache_array *array;
322 array = kmap(desc->page);
324 if (*desc->dir_cookie == 0)
325 status = nfs_readdir_search_for_pos(array, desc);
327 status = nfs_readdir_search_for_cookie(array, desc);
329 if (status == -EAGAIN) {
330 desc->last_cookie = array->last_cookie;
331 desc->current_index += array->size;
338 /* Fill a page with xdr information before transferring to the cache page */
340 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
341 struct nfs_entry *entry, struct file *file, struct inode *inode)
343 struct nfs_open_dir_context *ctx = file->private_data;
344 const struct cred *cred = ctx->cred;
345 unsigned long timestamp, gencount;
350 gencount = nfs_inc_attr_generation_counter();
351 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
352 NFS_SERVER(inode)->dtsize, desc->plus);
354 /* We requested READDIRPLUS, but the server doesn't grok it */
355 if (error == -ENOTSUPP && desc->plus) {
356 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
357 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
363 desc->timestamp = timestamp;
364 desc->gencount = gencount;
369 static int xdr_decode(nfs_readdir_descriptor_t *desc,
370 struct nfs_entry *entry, struct xdr_stream *xdr)
374 error = desc->decode(xdr, entry, desc->plus);
377 entry->fattr->time_start = desc->timestamp;
378 entry->fattr->gencount = desc->gencount;
382 /* Match file and dirent using either filehandle or fileid
383 * Note: caller is responsible for checking the fsid
386 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
389 struct nfs_inode *nfsi;
391 if (d_really_is_negative(dentry))
394 inode = d_inode(dentry);
395 if (is_bad_inode(inode) || NFS_STALE(inode))
399 if (entry->fattr->fileid != nfsi->fileid)
401 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
407 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
409 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
411 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
419 * This function is called by the lookup and getattr code to request the
420 * use of readdirplus to accelerate any future lookups in the same
423 void nfs_advise_use_readdirplus(struct inode *dir)
425 struct nfs_inode *nfsi = NFS_I(dir);
427 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
428 !list_empty(&nfsi->open_files))
429 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
433 * This function is mainly for use by nfs_getattr().
435 * If this is an 'ls -l', we want to force use of readdirplus.
436 * Do this by checking if there is an active file descriptor
437 * and calling nfs_advise_use_readdirplus, then forcing a
440 void nfs_force_use_readdirplus(struct inode *dir)
442 struct nfs_inode *nfsi = NFS_I(dir);
444 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
445 !list_empty(&nfsi->open_files)) {
446 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
447 invalidate_mapping_pages(dir->i_mapping, 0, -1);
452 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
454 struct qstr filename = QSTR_INIT(entry->name, entry->len);
455 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
456 struct dentry *dentry;
457 struct dentry *alias;
458 struct inode *dir = d_inode(parent);
462 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
464 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
466 if (filename.len == 0)
468 /* Validate that the name doesn't contain any illegal '\0' */
469 if (strnlen(filename.name, filename.len) != filename.len)
472 if (strnchr(filename.name, filename.len, '/'))
474 if (filename.name[0] == '.') {
475 if (filename.len == 1)
477 if (filename.len == 2 && filename.name[1] == '.')
480 filename.hash = full_name_hash(parent, filename.name, filename.len);
482 dentry = d_lookup(parent, &filename);
485 dentry = d_alloc_parallel(parent, &filename, &wq);
489 if (!d_in_lookup(dentry)) {
490 /* Is there a mountpoint here? If so, just exit */
491 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
492 &entry->fattr->fsid))
494 if (nfs_same_file(dentry, entry)) {
495 if (!entry->fh->size)
497 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
498 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
500 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
503 d_invalidate(dentry);
509 if (!entry->fh->size) {
510 d_lookup_done(dentry);
514 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
515 alias = d_splice_alias(inode, dentry);
516 d_lookup_done(dentry);
523 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
528 /* Perform conversion from xdr to cache array */
530 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
531 struct page **xdr_pages, struct page *page, unsigned int buflen)
533 struct xdr_stream stream;
535 struct page *scratch;
536 struct nfs_cache_array *array;
537 unsigned int count = 0;
539 int max_rapages = NFS_MAX_READDIR_RAPAGES;
541 desc->pvec.index = desc->page_index;
544 scratch = alloc_page(GFP_KERNEL);
551 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
552 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
555 status = xdr_decode(desc, entry, &stream);
557 if (status == -EAGAIN)
565 nfs_prime_dcache(file_dentry(desc->file), entry);
567 status = nfs_readdir_add_to_array(entry, desc->pvec.pages[desc->pvec.nr]);
568 if (status == -ENOSPC) {
570 if (desc->pvec.nr == max_rapages)
572 status = nfs_readdir_add_to_array(entry, desc->pvec.pages[desc->pvec.nr]);
576 } while (!entry->eof);
579 * page and desc->pvec.pages[0] are valid, don't need to check
580 * whether or not to be NULL.
582 copy_highpage(page, desc->pvec.pages[0]);
585 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
586 array = kmap_atomic(desc->pvec.pages[desc->pvec.nr]);
587 array->eof_index = array->size;
589 kunmap_atomic(array);
595 * desc->pvec.nr > 0 means at least one page was completely filled,
596 * we should return -ENOSPC. Otherwise function
597 * nfs_readdir_xdr_to_array will enter infinite loop.
599 if (desc->pvec.nr > 0)
605 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
608 for (i = 0; i < npages; i++)
613 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
614 * to nfs_readdir_free_pages()
617 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
621 for (i = 0; i < npages; i++) {
622 struct page *page = alloc_page(GFP_KERNEL);
630 nfs_readdir_free_pages(pages, i);
635 * nfs_readdir_rapages_init initialize rapages by nfs_cache_array structure.
638 void nfs_readdir_rapages_init(nfs_readdir_descriptor_t *desc)
640 struct nfs_cache_array *array;
641 int max_rapages = NFS_MAX_READDIR_RAPAGES;
644 for (index = 0; index < max_rapages; index++) {
645 array = kmap_atomic(desc->pvec.pages[index]);
646 memset(array, 0, sizeof(struct nfs_cache_array));
647 array->eof_index = -1;
648 kunmap_atomic(array);
653 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
655 struct page *pages[NFS_MAX_READDIR_PAGES];
656 struct nfs_entry entry;
657 struct file *file = desc->file;
658 struct nfs_cache_array *array;
659 int status = -ENOMEM;
660 unsigned int array_size = ARRAY_SIZE(pages);
663 * This means we hit readdir rdpages miss, the preallocated rdpages
664 * are useless, the preallocate rdpages should be reinitialized.
666 nfs_readdir_rapages_init(desc);
668 entry.prev_cookie = 0;
669 entry.cookie = desc->last_cookie;
671 entry.fh = nfs_alloc_fhandle();
672 entry.fattr = nfs_alloc_fattr();
673 entry.server = NFS_SERVER(inode);
674 if (entry.fh == NULL || entry.fattr == NULL)
677 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
678 if (IS_ERR(entry.label)) {
679 status = PTR_ERR(entry.label);
684 memset(array, 0, sizeof(struct nfs_cache_array));
685 array->eof_index = -1;
687 status = nfs_readdir_alloc_pages(pages, array_size);
689 goto out_release_array;
692 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
697 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
699 if (status == -ENOSPC)
703 } while (array->eof_index < 0);
705 nfs_readdir_free_pages(pages, array_size);
708 nfs4_label_free(entry.label);
710 nfs_free_fattr(entry.fattr);
711 nfs_free_fhandle(entry.fh);
716 * Now we cache directories properly, by converting xdr information
717 * to an array that can be used for lookups later. This results in
718 * fewer cache pages, since we can store more information on each page.
719 * We only need to convert from xdr once so future lookups are much simpler
722 int nfs_readdir_filler(void *data, struct page* page)
724 nfs_readdir_descriptor_t *desc = data;
725 struct inode *inode = file_inode(desc->file);
729 * If desc->page_index in range desc->pvec.index and
730 * desc->pvec.index + desc->pvec.nr, we get readdir cache hit.
732 if (desc->page_index >= desc->pvec.index &&
733 desc->page_index < (desc->pvec.index + desc->pvec.nr)) {
735 * page and desc->pvec.pages[x] are valid, don't need to check
736 * whether or not to be NULL.
738 copy_highpage(page, desc->pvec.pages[desc->page_index - desc->pvec.index]);
741 ret = nfs_readdir_xdr_to_array(desc, page, inode);
746 SetPageUptodate(page);
748 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
749 /* Should never happen */
750 nfs_zap_mapping(inode, inode->i_mapping);
760 void cache_page_release(nfs_readdir_descriptor_t *desc)
762 if (!desc->page->mapping)
763 nfs_readdir_clear_array(desc->page);
764 put_page(desc->page);
769 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
771 return read_cache_page(desc->file->f_mapping, desc->page_index,
772 nfs_readdir_filler, desc);
776 * Returns 0 if desc->dir_cookie was found on page desc->page_index
779 int find_cache_page(nfs_readdir_descriptor_t *desc)
783 desc->page = get_cache_page(desc);
784 if (IS_ERR(desc->page))
785 return PTR_ERR(desc->page);
787 res = nfs_readdir_search_array(desc);
789 cache_page_release(desc);
793 /* Search for desc->dir_cookie from the beginning of the page cache */
795 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
799 if (desc->page_index == 0) {
800 desc->current_index = 0;
801 desc->last_cookie = 0;
804 res = find_cache_page(desc);
805 } while (res == -EAGAIN);
810 * Once we've found the start of the dirent within a page: fill 'er up...
813 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
815 struct file *file = desc->file;
818 struct nfs_cache_array *array = NULL;
819 struct nfs_open_dir_context *ctx = file->private_data;
821 array = kmap(desc->page);
822 for (i = desc->cache_entry_index; i < array->size; i++) {
823 struct nfs_cache_array_entry *ent;
825 ent = &array->array[i];
826 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
827 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
832 if (i < (array->size-1))
833 *desc->dir_cookie = array->array[i+1].cookie;
835 *desc->dir_cookie = array->last_cookie;
839 if (array->eof_index >= 0)
843 cache_page_release(desc);
844 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
845 (unsigned long long)*desc->dir_cookie, res);
850 * If we cannot find a cookie in our cache, we suspect that this is
851 * because it points to a deleted file, so we ask the server to return
852 * whatever it thinks is the next entry. We then feed this to filldir.
853 * If all goes well, we should then be able to find our way round the
854 * cache on the next call to readdir_search_pagecache();
856 * NOTE: we cannot add the anonymous page to the pagecache because
857 * the data it contains might not be page aligned. Besides,
858 * we should already have a complete representation of the
859 * directory in the page cache by the time we get here.
862 int uncached_readdir(nfs_readdir_descriptor_t *desc)
864 struct page *page = NULL;
866 struct inode *inode = file_inode(desc->file);
867 struct nfs_open_dir_context *ctx = desc->file->private_data;
869 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
870 (unsigned long long)*desc->dir_cookie);
872 page = alloc_page(GFP_HIGHUSER);
878 desc->page_index = 0;
879 desc->last_cookie = *desc->dir_cookie;
883 status = nfs_readdir_xdr_to_array(desc, page, inode);
887 status = nfs_do_filldir(desc);
890 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
894 cache_page_release(desc);
898 /* The file offset position represents the dirent entry number. A
899 last cookie cache takes care of the common case of reading the
902 static int nfs_readdir(struct file *file, struct dir_context *ctx)
904 struct dentry *dentry = file_dentry(file);
905 struct inode *inode = d_inode(dentry);
906 nfs_readdir_descriptor_t my_desc,
908 struct nfs_open_dir_context *dir_ctx = file->private_data;
910 int max_rapages = NFS_MAX_READDIR_RAPAGES;
912 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
913 file, (long long)ctx->pos);
914 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
917 * ctx->pos points to the dirent entry number.
918 * *desc->dir_cookie has the cookie for the next entry. We have
919 * to either find the entry with the appropriate number or
920 * revalidate the cookie.
922 memset(desc, 0, sizeof(*desc));
926 desc->dir_cookie = &dir_ctx->dir_cookie;
927 desc->decode = NFS_PROTO(inode)->decode_dirent;
928 desc->plus = nfs_use_readdirplus(inode, ctx);
930 res = nfs_readdir_alloc_pages(desc->pvec.pages, max_rapages);
934 nfs_readdir_rapages_init(desc);
936 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
937 res = nfs_revalidate_mapping(inode, file->f_mapping);
942 res = readdir_search_pagecache(desc);
944 if (res == -EBADCOOKIE) {
946 /* This means either end of directory */
947 if (*desc->dir_cookie && !desc->eof) {
948 /* Or that the server has 'lost' a cookie */
949 res = uncached_readdir(desc);
955 if (res == -ETOOSMALL && desc->plus) {
956 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
957 nfs_zap_caches(inode);
958 desc->page_index = 0;
966 res = nfs_do_filldir(desc);
969 } while (!desc->eof);
971 nfs_readdir_free_pages(desc->pvec.pages, max_rapages);
974 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
978 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
980 struct inode *inode = file_inode(filp);
981 struct nfs_open_dir_context *dir_ctx = filp->private_data;
983 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
984 filp, offset, whence);
998 offset += filp->f_pos;
1000 inode_unlock(inode);
1004 if (offset != filp->f_pos) {
1005 filp->f_pos = offset;
1006 dir_ctx->dir_cookie = 0;
1009 inode_unlock(inode);
1014 * All directory operations under NFS are synchronous, so fsync()
1015 * is a dummy operation.
1017 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1020 struct inode *inode = file_inode(filp);
1022 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1025 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
1026 inode_unlock(inode);
1031 * nfs_force_lookup_revalidate - Mark the directory as having changed
1032 * @dir: pointer to directory inode
1034 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1035 * full lookup on all child dentries of 'dir' whenever a change occurs
1036 * on the server that might have invalidated our dcache.
1038 * The caller should be holding dir->i_lock
1040 void nfs_force_lookup_revalidate(struct inode *dir)
1042 NFS_I(dir)->cache_change_attribute++;
1044 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1047 * A check for whether or not the parent directory has changed.
1048 * In the case it has, we assume that the dentries are untrustworthy
1049 * and may need to be looked up again.
1050 * If rcu_walk prevents us from performing a full check, return 0.
1052 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1055 if (IS_ROOT(dentry))
1057 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1059 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1061 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1062 if (nfs_mapping_need_revalidate_inode(dir)) {
1065 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1068 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1074 * Use intent information to check whether or not we're going to do
1075 * an O_EXCL create using this path component.
1077 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1079 if (NFS_PROTO(dir)->version == 2)
1081 return flags & LOOKUP_EXCL;
1085 * Inode and filehandle revalidation for lookups.
1087 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1088 * or if the intent information indicates that we're about to open this
1089 * particular file and the "nocto" mount flag is not set.
1093 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1095 struct nfs_server *server = NFS_SERVER(inode);
1098 if (IS_AUTOMOUNT(inode))
1101 if (flags & LOOKUP_OPEN) {
1102 switch (inode->i_mode & S_IFMT) {
1104 /* A NFSv4 OPEN will revalidate later */
1105 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1109 if (server->flags & NFS_MOUNT_NOCTO)
1111 /* NFS close-to-open cache consistency validation */
1116 /* VFS wants an on-the-wire revalidation */
1117 if (flags & LOOKUP_REVAL)
1120 return (inode->i_nlink == 0) ? -ESTALE : 0;
1122 if (flags & LOOKUP_RCU)
1124 ret = __nfs_revalidate_inode(server, inode);
1131 * We judge how long we want to trust negative
1132 * dentries by looking at the parent inode mtime.
1134 * If parent mtime has changed, we revalidate, else we wait for a
1135 * period corresponding to the parent's attribute cache timeout value.
1137 * If LOOKUP_RCU prevents us from performing a full check, return 1
1138 * suggesting a reval is needed.
1140 * Note that when creating a new file, or looking up a rename target,
1141 * then it shouldn't be necessary to revalidate a negative dentry.
1144 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1147 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1149 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1151 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1155 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1156 struct inode *inode, int error)
1160 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1164 nfs_mark_for_revalidate(dir);
1165 if (inode && S_ISDIR(inode->i_mode)) {
1166 /* Purge readdir caches. */
1167 nfs_zap_caches(inode);
1169 * We can't d_drop the root of a disconnected tree:
1170 * its d_hash is on the s_anon list and d_drop() would hide
1171 * it from shrink_dcache_for_unmount(), leading to busy
1172 * inodes on unmount and further oopses.
1174 if (IS_ROOT(dentry))
1177 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1181 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1182 __func__, dentry, error);
1187 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1191 if (nfs_neg_need_reval(dir, dentry, flags)) {
1192 if (flags & LOOKUP_RCU)
1196 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1200 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1201 struct inode *inode)
1203 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1204 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1208 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1209 struct inode *inode)
1211 struct nfs_fh *fhandle;
1212 struct nfs_fattr *fattr;
1213 struct nfs4_label *label;
1217 fhandle = nfs_alloc_fhandle();
1218 fattr = nfs_alloc_fattr();
1219 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1220 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1223 ret = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1225 if (ret == -ESTALE || ret == -ENOENT)
1230 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1232 if (nfs_refresh_inode(inode, fattr) < 0)
1235 nfs_setsecurity(inode, fattr, label);
1236 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1238 /* set a readdirplus hint that we had a cache miss */
1239 nfs_force_use_readdirplus(dir);
1242 nfs_free_fattr(fattr);
1243 nfs_free_fhandle(fhandle);
1244 nfs4_label_free(label);
1245 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1249 * This is called every time the dcache has a lookup hit,
1250 * and we should check whether we can really trust that
1253 * NOTE! The hit can be a negative hit too, don't assume
1256 * If the parent directory is seen to have changed, we throw out the
1257 * cached dentry and do a new lookup.
1260 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1263 struct inode *inode;
1266 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1267 inode = d_inode(dentry);
1270 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1272 if (is_bad_inode(inode)) {
1273 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1278 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1279 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1281 /* Force a full look up iff the parent directory has changed */
1282 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1283 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1284 error = nfs_lookup_verify_inode(inode, flags);
1286 if (error == -ESTALE)
1287 nfs_zap_caches(dir);
1290 nfs_advise_use_readdirplus(dir);
1294 if (flags & LOOKUP_RCU)
1297 if (NFS_STALE(inode))
1300 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1301 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1302 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1305 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1307 if (flags & LOOKUP_RCU)
1309 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1313 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1314 int (*reval)(struct inode *, struct dentry *, unsigned int))
1316 struct dentry *parent;
1320 if (flags & LOOKUP_RCU) {
1321 parent = READ_ONCE(dentry->d_parent);
1322 dir = d_inode_rcu(parent);
1325 ret = reval(dir, dentry, flags);
1326 if (parent != READ_ONCE(dentry->d_parent))
1329 parent = dget_parent(dentry);
1330 ret = reval(d_inode(parent), dentry, flags);
1336 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1338 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1342 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1343 * when we don't really care about the dentry name. This is called when a
1344 * pathwalk ends on a dentry that was not found via a normal lookup in the
1345 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1347 * In this situation, we just want to verify that the inode itself is OK
1348 * since the dentry might have changed on the server.
1350 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1352 struct inode *inode = d_inode(dentry);
1356 * I believe we can only get a negative dentry here in the case of a
1357 * procfs-style symlink. Just assume it's correct for now, but we may
1358 * eventually need to do something more here.
1361 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1366 if (is_bad_inode(inode)) {
1367 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1372 error = nfs_lookup_verify_inode(inode, flags);
1373 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1374 __func__, inode->i_ino, error ? "invalid" : "valid");
1379 * This is called from dput() when d_count is going to 0.
1381 static int nfs_dentry_delete(const struct dentry *dentry)
1383 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1384 dentry, dentry->d_flags);
1386 /* Unhash any dentry with a stale inode */
1387 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1390 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1391 /* Unhash it, so that ->d_iput() would be called */
1394 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1395 /* Unhash it, so that ancestors of killed async unlink
1396 * files will be cleaned up during umount */
1403 /* Ensure that we revalidate inode->i_nlink */
1404 static void nfs_drop_nlink(struct inode *inode)
1406 spin_lock(&inode->i_lock);
1407 /* drop the inode if we're reasonably sure this is the last link */
1408 if (inode->i_nlink > 0)
1410 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1411 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1412 | NFS_INO_INVALID_CTIME
1413 | NFS_INO_INVALID_OTHER
1414 | NFS_INO_REVAL_FORCED;
1415 spin_unlock(&inode->i_lock);
1419 * Called when the dentry loses inode.
1420 * We use it to clean up silly-renamed files.
1422 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1424 if (S_ISDIR(inode->i_mode))
1425 /* drop any readdir cache as it could easily be old */
1426 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1428 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1429 nfs_complete_unlink(dentry, inode);
1430 nfs_drop_nlink(inode);
1435 static void nfs_d_release(struct dentry *dentry)
1437 /* free cached devname value, if it survived that far */
1438 if (unlikely(dentry->d_fsdata)) {
1439 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1442 kfree(dentry->d_fsdata);
1446 const struct dentry_operations nfs_dentry_operations = {
1447 .d_revalidate = nfs_lookup_revalidate,
1448 .d_weak_revalidate = nfs_weak_revalidate,
1449 .d_delete = nfs_dentry_delete,
1450 .d_iput = nfs_dentry_iput,
1451 .d_automount = nfs_d_automount,
1452 .d_release = nfs_d_release,
1454 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1456 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1459 struct inode *inode = NULL;
1460 struct nfs_fh *fhandle = NULL;
1461 struct nfs_fattr *fattr = NULL;
1462 struct nfs4_label *label = NULL;
1465 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1466 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1468 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1469 return ERR_PTR(-ENAMETOOLONG);
1472 * If we're doing an exclusive create, optimize away the lookup
1473 * but don't hash the dentry.
1475 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1478 res = ERR_PTR(-ENOMEM);
1479 fhandle = nfs_alloc_fhandle();
1480 fattr = nfs_alloc_fattr();
1481 if (fhandle == NULL || fattr == NULL)
1484 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1488 trace_nfs_lookup_enter(dir, dentry, flags);
1489 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1490 if (error == -ENOENT)
1493 res = ERR_PTR(error);
1496 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1497 res = ERR_CAST(inode);
1501 /* Notify readdir to use READDIRPLUS */
1502 nfs_force_use_readdirplus(dir);
1505 res = d_splice_alias(inode, dentry);
1511 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1513 trace_nfs_lookup_exit(dir, dentry, flags, error);
1514 nfs4_label_free(label);
1516 nfs_free_fattr(fattr);
1517 nfs_free_fhandle(fhandle);
1520 EXPORT_SYMBOL_GPL(nfs_lookup);
1522 #if IS_ENABLED(CONFIG_NFS_V4)
1523 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1525 const struct dentry_operations nfs4_dentry_operations = {
1526 .d_revalidate = nfs4_lookup_revalidate,
1527 .d_weak_revalidate = nfs_weak_revalidate,
1528 .d_delete = nfs_dentry_delete,
1529 .d_iput = nfs_dentry_iput,
1530 .d_automount = nfs_d_automount,
1531 .d_release = nfs_d_release,
1533 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1535 static fmode_t flags_to_mode(int flags)
1537 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1538 if ((flags & O_ACCMODE) != O_WRONLY)
1540 if ((flags & O_ACCMODE) != O_RDONLY)
1545 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1547 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1550 static int do_open(struct inode *inode, struct file *filp)
1552 nfs_fscache_open_file(inode, filp);
1556 static int nfs_finish_open(struct nfs_open_context *ctx,
1557 struct dentry *dentry,
1558 struct file *file, unsigned open_flags)
1562 err = finish_open(file, dentry, do_open);
1565 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1566 nfs_file_set_open_context(file, ctx);
1573 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1574 struct file *file, unsigned open_flags,
1577 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1578 struct nfs_open_context *ctx;
1580 struct iattr attr = { .ia_valid = ATTR_OPEN };
1581 struct inode *inode;
1582 unsigned int lookup_flags = 0;
1583 bool switched = false;
1587 /* Expect a negative dentry */
1588 BUG_ON(d_inode(dentry));
1590 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1591 dir->i_sb->s_id, dir->i_ino, dentry);
1593 err = nfs_check_flags(open_flags);
1597 /* NFS only supports OPEN on regular files */
1598 if ((open_flags & O_DIRECTORY)) {
1599 if (!d_in_lookup(dentry)) {
1601 * Hashed negative dentry with O_DIRECTORY: dentry was
1602 * revalidated and is fine, no need to perform lookup
1607 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1611 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1612 return -ENAMETOOLONG;
1614 if (open_flags & O_CREAT) {
1615 struct nfs_server *server = NFS_SERVER(dir);
1617 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1618 mode &= ~current_umask();
1620 attr.ia_valid |= ATTR_MODE;
1621 attr.ia_mode = mode;
1623 if (open_flags & O_TRUNC) {
1624 attr.ia_valid |= ATTR_SIZE;
1628 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1631 dentry = d_alloc_parallel(dentry->d_parent,
1632 &dentry->d_name, &wq);
1634 return PTR_ERR(dentry);
1635 if (unlikely(!d_in_lookup(dentry)))
1636 return finish_no_open(file, dentry);
1639 ctx = create_nfs_open_context(dentry, open_flags, file);
1644 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1645 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1647 file->f_mode |= FMODE_CREATED;
1648 if (IS_ERR(inode)) {
1649 err = PTR_ERR(inode);
1650 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1651 put_nfs_open_context(ctx);
1655 d_splice_alias(NULL, dentry);
1656 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1662 if (!(open_flags & O_NOFOLLOW))
1672 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1673 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1674 put_nfs_open_context(ctx);
1676 if (unlikely(switched)) {
1677 d_lookup_done(dentry);
1683 res = nfs_lookup(dir, dentry, lookup_flags);
1685 d_lookup_done(dentry);
1692 return PTR_ERR(res);
1693 return finish_no_open(file, res);
1695 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1698 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1701 struct inode *inode;
1703 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1705 if (d_mountpoint(dentry))
1708 inode = d_inode(dentry);
1710 /* We can't create new files in nfs_open_revalidate(), so we
1711 * optimize away revalidation of negative dentries.
1716 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1717 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1719 /* NFS only supports OPEN on regular files */
1720 if (!S_ISREG(inode->i_mode))
1723 /* We cannot do exclusive creation on a positive dentry */
1724 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1727 /* Check if the directory changed */
1728 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1731 /* Let f_op->open() actually open (and revalidate) the file */
1734 if (flags & LOOKUP_RCU)
1736 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1739 return nfs_do_lookup_revalidate(dir, dentry, flags);
1742 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1744 return __nfs_lookup_revalidate(dentry, flags,
1745 nfs4_do_lookup_revalidate);
1748 #endif /* CONFIG_NFSV4 */
1751 * Code common to create, mkdir, and mknod.
1753 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1754 struct nfs_fattr *fattr,
1755 struct nfs4_label *label)
1757 struct dentry *parent = dget_parent(dentry);
1758 struct inode *dir = d_inode(parent);
1759 struct inode *inode;
1761 int error = -EACCES;
1765 /* We may have been initialized further down */
1766 if (d_really_is_positive(dentry))
1768 if (fhandle->size == 0) {
1769 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1773 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1774 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1775 struct nfs_server *server = NFS_SB(dentry->d_sb);
1776 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1781 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1782 d = d_splice_alias(inode, dentry);
1792 nfs_mark_for_revalidate(dir);
1796 EXPORT_SYMBOL_GPL(nfs_instantiate);
1799 * Following a failed create operation, we drop the dentry rather
1800 * than retain a negative dentry. This avoids a problem in the event
1801 * that the operation succeeded on the server, but an error in the
1802 * reply path made it appear to have failed.
1804 int nfs_create(struct inode *dir, struct dentry *dentry,
1805 umode_t mode, bool excl)
1808 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1811 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1812 dir->i_sb->s_id, dir->i_ino, dentry);
1814 attr.ia_mode = mode;
1815 attr.ia_valid = ATTR_MODE;
1817 trace_nfs_create_enter(dir, dentry, open_flags);
1818 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1819 trace_nfs_create_exit(dir, dentry, open_flags, error);
1827 EXPORT_SYMBOL_GPL(nfs_create);
1830 * See comments for nfs_proc_create regarding failed operations.
1833 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1838 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1839 dir->i_sb->s_id, dir->i_ino, dentry);
1841 attr.ia_mode = mode;
1842 attr.ia_valid = ATTR_MODE;
1844 trace_nfs_mknod_enter(dir, dentry);
1845 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1846 trace_nfs_mknod_exit(dir, dentry, status);
1854 EXPORT_SYMBOL_GPL(nfs_mknod);
1857 * See comments for nfs_proc_create regarding failed operations.
1859 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1864 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1865 dir->i_sb->s_id, dir->i_ino, dentry);
1867 attr.ia_valid = ATTR_MODE;
1868 attr.ia_mode = mode | S_IFDIR;
1870 trace_nfs_mkdir_enter(dir, dentry);
1871 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1872 trace_nfs_mkdir_exit(dir, dentry, error);
1880 EXPORT_SYMBOL_GPL(nfs_mkdir);
1882 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1884 if (simple_positive(dentry))
1888 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1892 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1893 dir->i_sb->s_id, dir->i_ino, dentry);
1895 trace_nfs_rmdir_enter(dir, dentry);
1896 if (d_really_is_positive(dentry)) {
1897 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1898 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1899 /* Ensure the VFS deletes this inode */
1902 clear_nlink(d_inode(dentry));
1905 nfs_dentry_handle_enoent(dentry);
1907 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1909 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1910 trace_nfs_rmdir_exit(dir, dentry, error);
1914 EXPORT_SYMBOL_GPL(nfs_rmdir);
1917 * Remove a file after making sure there are no pending writes,
1918 * and after checking that the file has only one user.
1920 * We invalidate the attribute cache and free the inode prior to the operation
1921 * to avoid possible races if the server reuses the inode.
1923 static int nfs_safe_remove(struct dentry *dentry)
1925 struct inode *dir = d_inode(dentry->d_parent);
1926 struct inode *inode = d_inode(dentry);
1929 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1931 /* If the dentry was sillyrenamed, we simply call d_delete() */
1932 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1937 trace_nfs_remove_enter(dir, dentry);
1938 if (inode != NULL) {
1939 error = NFS_PROTO(dir)->remove(dir, dentry);
1941 nfs_drop_nlink(inode);
1943 error = NFS_PROTO(dir)->remove(dir, dentry);
1944 if (error == -ENOENT)
1945 nfs_dentry_handle_enoent(dentry);
1946 trace_nfs_remove_exit(dir, dentry, error);
1951 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1952 * belongs to an active ".nfs..." file and we return -EBUSY.
1954 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1956 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1959 int need_rehash = 0;
1961 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1962 dir->i_ino, dentry);
1964 trace_nfs_unlink_enter(dir, dentry);
1965 spin_lock(&dentry->d_lock);
1966 if (d_count(dentry) > 1) {
1967 spin_unlock(&dentry->d_lock);
1968 /* Start asynchronous writeout of the inode */
1969 write_inode_now(d_inode(dentry), 0);
1970 error = nfs_sillyrename(dir, dentry);
1973 if (!d_unhashed(dentry)) {
1977 spin_unlock(&dentry->d_lock);
1978 error = nfs_safe_remove(dentry);
1979 if (!error || error == -ENOENT) {
1980 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1981 } else if (need_rehash)
1984 trace_nfs_unlink_exit(dir, dentry, error);
1987 EXPORT_SYMBOL_GPL(nfs_unlink);
1990 * To create a symbolic link, most file systems instantiate a new inode,
1991 * add a page to it containing the path, then write it out to the disk
1992 * using prepare_write/commit_write.
1994 * Unfortunately the NFS client can't create the in-core inode first
1995 * because it needs a file handle to create an in-core inode (see
1996 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1997 * symlink request has completed on the server.
1999 * So instead we allocate a raw page, copy the symname into it, then do
2000 * the SYMLINK request with the page as the buffer. If it succeeds, we
2001 * now have a new file handle and can instantiate an in-core NFS inode
2002 * and move the raw page into its mapping.
2004 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2009 unsigned int pathlen = strlen(symname);
2012 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2013 dir->i_ino, dentry, symname);
2015 if (pathlen > PAGE_SIZE)
2016 return -ENAMETOOLONG;
2018 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2019 attr.ia_valid = ATTR_MODE;
2021 page = alloc_page(GFP_USER);
2025 kaddr = page_address(page);
2026 memcpy(kaddr, symname, pathlen);
2027 if (pathlen < PAGE_SIZE)
2028 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2030 trace_nfs_symlink_enter(dir, dentry);
2031 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2032 trace_nfs_symlink_exit(dir, dentry, error);
2034 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2035 dir->i_sb->s_id, dir->i_ino,
2036 dentry, symname, error);
2043 * No big deal if we can't add this page to the page cache here.
2044 * READLINK will get the missing page from the server if needed.
2046 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2048 SetPageUptodate(page);
2051 * add_to_page_cache_lru() grabs an extra page refcount.
2052 * Drop it here to avoid leaking this page later.
2060 EXPORT_SYMBOL_GPL(nfs_symlink);
2063 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2065 struct inode *inode = d_inode(old_dentry);
2068 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2069 old_dentry, dentry);
2071 trace_nfs_link_enter(inode, dir, dentry);
2073 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2076 d_add(dentry, inode);
2078 trace_nfs_link_exit(inode, dir, dentry, error);
2081 EXPORT_SYMBOL_GPL(nfs_link);
2085 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2086 * different file handle for the same inode after a rename (e.g. when
2087 * moving to a different directory). A fail-safe method to do so would
2088 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2089 * rename the old file using the sillyrename stuff. This way, the original
2090 * file in old_dir will go away when the last process iput()s the inode.
2094 * It actually works quite well. One needs to have the possibility for
2095 * at least one ".nfs..." file in each directory the file ever gets
2096 * moved or linked to which happens automagically with the new
2097 * implementation that only depends on the dcache stuff instead of
2098 * using the inode layer
2100 * Unfortunately, things are a little more complicated than indicated
2101 * above. For a cross-directory move, we want to make sure we can get
2102 * rid of the old inode after the operation. This means there must be
2103 * no pending writes (if it's a file), and the use count must be 1.
2104 * If these conditions are met, we can drop the dentries before doing
2107 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2108 struct inode *new_dir, struct dentry *new_dentry,
2111 struct inode *old_inode = d_inode(old_dentry);
2112 struct inode *new_inode = d_inode(new_dentry);
2113 struct dentry *dentry = NULL, *rehash = NULL;
2114 struct rpc_task *task;
2120 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2121 old_dentry, new_dentry,
2122 d_count(new_dentry));
2124 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2126 * For non-directories, check whether the target is busy and if so,
2127 * make a copy of the dentry and then do a silly-rename. If the
2128 * silly-rename succeeds, the copied dentry is hashed and becomes
2131 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2133 * To prevent any new references to the target during the
2134 * rename, we unhash the dentry in advance.
2136 if (!d_unhashed(new_dentry)) {
2138 rehash = new_dentry;
2141 if (d_count(new_dentry) > 2) {
2144 /* copy the target dentry's name */
2145 dentry = d_alloc(new_dentry->d_parent,
2146 &new_dentry->d_name);
2150 /* silly-rename the existing target ... */
2151 err = nfs_sillyrename(new_dir, new_dentry);
2155 new_dentry = dentry;
2161 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2163 error = PTR_ERR(task);
2167 error = rpc_wait_for_completion_task(task);
2169 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2170 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2173 error = task->tk_status;
2175 /* Ensure the inode attributes are revalidated */
2177 spin_lock(&old_inode->i_lock);
2178 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2179 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2180 | NFS_INO_INVALID_CTIME
2181 | NFS_INO_REVAL_FORCED;
2182 spin_unlock(&old_inode->i_lock);
2187 trace_nfs_rename_exit(old_dir, old_dentry,
2188 new_dir, new_dentry, error);
2190 if (new_inode != NULL)
2191 nfs_drop_nlink(new_inode);
2193 * The d_move() should be here instead of in an async RPC completion
2194 * handler because we need the proper locks to move the dentry. If
2195 * we're interrupted by a signal, the async RPC completion handler
2196 * should mark the directories for revalidation.
2198 d_move(old_dentry, new_dentry);
2199 nfs_set_verifier(old_dentry,
2200 nfs_save_change_attribute(new_dir));
2201 } else if (error == -ENOENT)
2202 nfs_dentry_handle_enoent(old_dentry);
2204 /* new dentry created? */
2209 EXPORT_SYMBOL_GPL(nfs_rename);
2211 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2212 static LIST_HEAD(nfs_access_lru_list);
2213 static atomic_long_t nfs_access_nr_entries;
2215 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2216 module_param(nfs_access_max_cachesize, ulong, 0644);
2217 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2219 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2221 put_cred(entry->cred);
2222 kfree_rcu(entry, rcu_head);
2223 smp_mb__before_atomic();
2224 atomic_long_dec(&nfs_access_nr_entries);
2225 smp_mb__after_atomic();
2228 static void nfs_access_free_list(struct list_head *head)
2230 struct nfs_access_entry *cache;
2232 while (!list_empty(head)) {
2233 cache = list_entry(head->next, struct nfs_access_entry, lru);
2234 list_del(&cache->lru);
2235 nfs_access_free_entry(cache);
2239 static unsigned long
2240 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2243 struct nfs_inode *nfsi, *next;
2244 struct nfs_access_entry *cache;
2247 spin_lock(&nfs_access_lru_lock);
2248 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2249 struct inode *inode;
2251 if (nr_to_scan-- == 0)
2253 inode = &nfsi->vfs_inode;
2254 spin_lock(&inode->i_lock);
2255 if (list_empty(&nfsi->access_cache_entry_lru))
2256 goto remove_lru_entry;
2257 cache = list_entry(nfsi->access_cache_entry_lru.next,
2258 struct nfs_access_entry, lru);
2259 list_move(&cache->lru, &head);
2260 rb_erase(&cache->rb_node, &nfsi->access_cache);
2262 if (!list_empty(&nfsi->access_cache_entry_lru))
2263 list_move_tail(&nfsi->access_cache_inode_lru,
2264 &nfs_access_lru_list);
2267 list_del_init(&nfsi->access_cache_inode_lru);
2268 smp_mb__before_atomic();
2269 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2270 smp_mb__after_atomic();
2272 spin_unlock(&inode->i_lock);
2274 spin_unlock(&nfs_access_lru_lock);
2275 nfs_access_free_list(&head);
2280 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2282 int nr_to_scan = sc->nr_to_scan;
2283 gfp_t gfp_mask = sc->gfp_mask;
2285 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2287 return nfs_do_access_cache_scan(nr_to_scan);
2292 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2294 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2298 nfs_access_cache_enforce_limit(void)
2300 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2302 unsigned int nr_to_scan;
2304 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2307 diff = nr_entries - nfs_access_max_cachesize;
2308 if (diff < nr_to_scan)
2310 nfs_do_access_cache_scan(nr_to_scan);
2313 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2315 struct rb_root *root_node = &nfsi->access_cache;
2317 struct nfs_access_entry *entry;
2319 /* Unhook entries from the cache */
2320 while ((n = rb_first(root_node)) != NULL) {
2321 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2322 rb_erase(n, root_node);
2323 list_move(&entry->lru, head);
2325 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2328 void nfs_access_zap_cache(struct inode *inode)
2332 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2334 /* Remove from global LRU init */
2335 spin_lock(&nfs_access_lru_lock);
2336 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2337 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2339 spin_lock(&inode->i_lock);
2340 __nfs_access_zap_cache(NFS_I(inode), &head);
2341 spin_unlock(&inode->i_lock);
2342 spin_unlock(&nfs_access_lru_lock);
2343 nfs_access_free_list(&head);
2345 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2347 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2349 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2352 struct nfs_access_entry *entry =
2353 rb_entry(n, struct nfs_access_entry, rb_node);
2354 int cmp = cred_fscmp(cred, entry->cred);
2366 static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2368 struct nfs_inode *nfsi = NFS_I(inode);
2369 struct nfs_access_entry *cache;
2373 spin_lock(&inode->i_lock);
2375 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2377 cache = nfs_access_search_rbtree(inode, cred);
2381 /* Found an entry, is our attribute cache valid? */
2382 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2389 spin_unlock(&inode->i_lock);
2390 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2393 spin_lock(&inode->i_lock);
2396 res->cred = cache->cred;
2397 res->mask = cache->mask;
2398 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2401 spin_unlock(&inode->i_lock);
2404 spin_unlock(&inode->i_lock);
2405 nfs_access_zap_cache(inode);
2409 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2411 /* Only check the most recently returned cache entry,
2412 * but do it without locking.
2414 struct nfs_inode *nfsi = NFS_I(inode);
2415 struct nfs_access_entry *cache;
2417 struct list_head *lh;
2420 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2422 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2423 cache = list_entry(lh, struct nfs_access_entry, lru);
2424 if (lh == &nfsi->access_cache_entry_lru ||
2425 cred != cache->cred)
2429 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2431 res->cred = cache->cred;
2432 res->mask = cache->mask;
2439 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2441 struct nfs_inode *nfsi = NFS_I(inode);
2442 struct rb_root *root_node = &nfsi->access_cache;
2443 struct rb_node **p = &root_node->rb_node;
2444 struct rb_node *parent = NULL;
2445 struct nfs_access_entry *entry;
2448 spin_lock(&inode->i_lock);
2449 while (*p != NULL) {
2451 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2452 cmp = cred_fscmp(set->cred, entry->cred);
2455 p = &parent->rb_left;
2457 p = &parent->rb_right;
2461 rb_link_node(&set->rb_node, parent, p);
2462 rb_insert_color(&set->rb_node, root_node);
2463 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2464 spin_unlock(&inode->i_lock);
2467 rb_replace_node(parent, &set->rb_node, root_node);
2468 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2469 list_del(&entry->lru);
2470 spin_unlock(&inode->i_lock);
2471 nfs_access_free_entry(entry);
2474 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2476 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2479 RB_CLEAR_NODE(&cache->rb_node);
2480 cache->cred = get_cred(set->cred);
2481 cache->mask = set->mask;
2483 /* The above field assignments must be visible
2484 * before this item appears on the lru. We cannot easily
2485 * use rcu_assign_pointer, so just force the memory barrier.
2488 nfs_access_add_rbtree(inode, cache);
2490 /* Update accounting */
2491 smp_mb__before_atomic();
2492 atomic_long_inc(&nfs_access_nr_entries);
2493 smp_mb__after_atomic();
2495 /* Add inode to global LRU list */
2496 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2497 spin_lock(&nfs_access_lru_lock);
2498 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2499 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2500 &nfs_access_lru_list);
2501 spin_unlock(&nfs_access_lru_lock);
2503 nfs_access_cache_enforce_limit();
2505 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2507 #define NFS_MAY_READ (NFS_ACCESS_READ)
2508 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2509 NFS_ACCESS_EXTEND | \
2511 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2513 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2514 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2515 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2517 nfs_access_calc_mask(u32 access_result, umode_t umode)
2521 if (access_result & NFS_MAY_READ)
2523 if (S_ISDIR(umode)) {
2524 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2526 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2528 } else if (S_ISREG(umode)) {
2529 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2531 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2533 } else if (access_result & NFS_MAY_WRITE)
2538 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2540 entry->mask = access_result;
2542 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2544 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2546 struct nfs_access_entry cache;
2547 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2551 trace_nfs_access_enter(inode);
2553 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2555 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2564 * Determine which access bits we want to ask for...
2566 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2567 if (S_ISDIR(inode->i_mode))
2568 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2570 cache.mask |= NFS_ACCESS_EXECUTE;
2572 status = NFS_PROTO(inode)->access(inode, &cache);
2574 if (status == -ESTALE) {
2575 nfs_zap_caches(inode);
2576 if (!S_ISDIR(inode->i_mode))
2577 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2581 nfs_access_add_cache(inode, &cache);
2583 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2584 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2587 trace_nfs_access_exit(inode, status);
2591 static int nfs_open_permission_mask(int openflags)
2595 if (openflags & __FMODE_EXEC) {
2596 /* ONLY check exec rights */
2599 if ((openflags & O_ACCMODE) != O_WRONLY)
2601 if ((openflags & O_ACCMODE) != O_RDONLY)
2608 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2610 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2612 EXPORT_SYMBOL_GPL(nfs_may_open);
2614 static int nfs_execute_ok(struct inode *inode, int mask)
2616 struct nfs_server *server = NFS_SERVER(inode);
2619 if (S_ISDIR(inode->i_mode))
2621 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2622 if (mask & MAY_NOT_BLOCK)
2624 ret = __nfs_revalidate_inode(server, inode);
2626 if (ret == 0 && !execute_ok(inode))
2631 int nfs_permission(struct inode *inode, int mask)
2633 const struct cred *cred = current_cred();
2636 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2638 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2640 /* Is this sys_access() ? */
2641 if (mask & (MAY_ACCESS | MAY_CHDIR))
2644 switch (inode->i_mode & S_IFMT) {
2648 if ((mask & MAY_OPEN) &&
2649 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2654 * Optimize away all write operations, since the server
2655 * will check permissions when we perform the op.
2657 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2662 if (!NFS_PROTO(inode)->access)
2665 /* Always try fast lookups first */
2667 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2669 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2670 /* Fast lookup failed, try the slow way */
2671 res = nfs_do_access(inode, cred, mask);
2674 if (!res && (mask & MAY_EXEC))
2675 res = nfs_execute_ok(inode, mask);
2677 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2678 inode->i_sb->s_id, inode->i_ino, mask, res);
2681 if (mask & MAY_NOT_BLOCK)
2684 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2686 res = generic_permission(inode, mask);
2689 EXPORT_SYMBOL_GPL(nfs_permission);
2693 * version-control: t
2694 * kept-new-versions: 5