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_shared = 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 {
143 unsigned char page_full : 1,
145 struct nfs_cache_array_entry array[];
148 typedef struct nfs_readdir_descriptor {
151 struct dir_context *ctx;
152 unsigned long page_index;
156 loff_t current_index;
159 unsigned long dir_verifier;
160 unsigned long timestamp;
161 unsigned long gencount;
162 unsigned long attr_gencount;
163 unsigned int cache_entry_index;
167 } nfs_readdir_descriptor_t;
170 void nfs_readdir_init_array(struct page *page)
172 struct nfs_cache_array *array;
174 array = kmap_atomic(page);
175 memset(array, 0, sizeof(struct nfs_cache_array));
176 kunmap_atomic(array);
180 * we are freeing strings created by nfs_add_to_readdir_array()
183 void nfs_readdir_clear_array(struct page *page)
185 struct nfs_cache_array *array;
188 array = kmap_atomic(page);
189 for (i = 0; i < array->size; i++)
190 kfree(array->array[i].string.name);
192 kunmap_atomic(array);
195 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
197 array->page_is_eof = 1;
198 array->page_full = 1;
201 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
203 return array->page_full;
207 * the caller is responsible for freeing qstr.name
208 * when called by nfs_readdir_add_to_array, the strings will be freed in
209 * nfs_clear_readdir_array()
212 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
215 string->name = kmemdup_nul(name, len, GFP_KERNEL);
216 if (string->name == NULL)
219 * Avoid a kmemleak false positive. The pointer to the name is stored
220 * in a page cache page which kmemleak does not scan.
222 kmemleak_not_leak(string->name);
223 string->hash = full_name_hash(NULL, name, len);
228 * Check that the next array entry lies entirely within the page bounds
230 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
232 struct nfs_cache_array_entry *cache_entry;
234 if (array->page_full)
236 cache_entry = &array->array[array->size + 1];
237 if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
238 array->page_full = 1;
245 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
247 struct nfs_cache_array *array = kmap(page);
248 struct nfs_cache_array_entry *cache_entry;
251 ret = nfs_readdir_array_can_expand(array);
255 cache_entry = &array->array[array->size];
256 cache_entry->cookie = entry->prev_cookie;
257 cache_entry->ino = entry->ino;
258 cache_entry->d_type = entry->d_type;
259 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
262 array->last_cookie = entry->cookie;
265 nfs_readdir_array_set_eof(array);
271 static void nfs_readdir_page_set_eof(struct page *page)
273 struct nfs_cache_array *array;
275 array = kmap_atomic(page);
276 nfs_readdir_array_set_eof(array);
277 kunmap_atomic(array);
281 int is_32bit_api(void)
284 return in_compat_syscall();
286 return (BITS_PER_LONG == 32);
291 bool nfs_readdir_use_cookie(const struct file *filp)
293 if ((filp->f_mode & FMODE_32BITHASH) ||
294 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
300 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
302 loff_t diff = desc->ctx->pos - desc->current_index;
307 if (diff >= array->size) {
308 if (array->page_is_eof)
313 index = (unsigned int)diff;
314 desc->dir_cookie = array->array[index].cookie;
315 desc->cache_entry_index = index;
323 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
325 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
328 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
332 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
336 int status = -EAGAIN;
338 for (i = 0; i < array->size; i++) {
339 if (array->array[i].cookie == desc->dir_cookie) {
340 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
342 new_pos = desc->current_index + i;
343 if (desc->attr_gencount != nfsi->attr_gencount ||
344 !nfs_readdir_inode_mapping_valid(nfsi)) {
346 desc->attr_gencount = nfsi->attr_gencount;
347 } else if (new_pos < desc->prev_index) {
349 && desc->dup_cookie == desc->dir_cookie) {
350 if (printk_ratelimit()) {
351 pr_notice("NFS: directory %pD2 contains a readdir loop."
352 "Please contact your server vendor. "
353 "The file: %.*s has duplicate cookie %llu\n",
354 desc->file, array->array[i].string.len,
355 array->array[i].string.name, desc->dir_cookie);
360 desc->dup_cookie = desc->dir_cookie;
363 if (nfs_readdir_use_cookie(desc->file))
364 desc->ctx->pos = desc->dir_cookie;
366 desc->ctx->pos = new_pos;
367 desc->prev_index = new_pos;
368 desc->cache_entry_index = i;
372 if (array->page_is_eof) {
373 status = -EBADCOOKIE;
374 if (desc->dir_cookie == array->last_cookie)
382 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
384 struct nfs_cache_array *array;
387 array = kmap(desc->page);
389 if (desc->dir_cookie == 0)
390 status = nfs_readdir_search_for_pos(array, desc);
392 status = nfs_readdir_search_for_cookie(array, desc);
394 if (status == -EAGAIN) {
395 desc->last_cookie = array->last_cookie;
396 desc->current_index += array->size;
403 /* Fill a page with xdr information before transferring to the cache page */
405 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
406 struct nfs_entry *entry, struct file *file, struct inode *inode)
408 struct nfs_open_dir_context *ctx = file->private_data;
409 const struct cred *cred = ctx->cred;
410 unsigned long timestamp, gencount;
415 gencount = nfs_inc_attr_generation_counter();
416 desc->dir_verifier = nfs_save_change_attribute(inode);
417 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
418 NFS_SERVER(inode)->dtsize, desc->plus);
420 /* We requested READDIRPLUS, but the server doesn't grok it */
421 if (error == -ENOTSUPP && desc->plus) {
422 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
423 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
429 desc->timestamp = timestamp;
430 desc->gencount = gencount;
435 static int xdr_decode(nfs_readdir_descriptor_t *desc,
436 struct nfs_entry *entry, struct xdr_stream *xdr)
438 struct inode *inode = file_inode(desc->file);
441 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
444 entry->fattr->time_start = desc->timestamp;
445 entry->fattr->gencount = desc->gencount;
449 /* Match file and dirent using either filehandle or fileid
450 * Note: caller is responsible for checking the fsid
453 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
456 struct nfs_inode *nfsi;
458 if (d_really_is_negative(dentry))
461 inode = d_inode(dentry);
462 if (is_bad_inode(inode) || NFS_STALE(inode))
466 if (entry->fattr->fileid != nfsi->fileid)
468 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
474 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
476 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
478 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
486 * This function is called by the lookup and getattr code to request the
487 * use of readdirplus to accelerate any future lookups in the same
490 void nfs_advise_use_readdirplus(struct inode *dir)
492 struct nfs_inode *nfsi = NFS_I(dir);
494 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
495 !list_empty(&nfsi->open_files))
496 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
500 * This function is mainly for use by nfs_getattr().
502 * If this is an 'ls -l', we want to force use of readdirplus.
503 * Do this by checking if there is an active file descriptor
504 * and calling nfs_advise_use_readdirplus, then forcing a
507 void nfs_force_use_readdirplus(struct inode *dir)
509 struct nfs_inode *nfsi = NFS_I(dir);
511 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
512 !list_empty(&nfsi->open_files)) {
513 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
514 invalidate_mapping_pages(dir->i_mapping,
515 nfsi->page_index + 1, -1);
520 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
521 unsigned long dir_verifier)
523 struct qstr filename = QSTR_INIT(entry->name, entry->len);
524 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
525 struct dentry *dentry;
526 struct dentry *alias;
530 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
532 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
534 if (filename.len == 0)
536 /* Validate that the name doesn't contain any illegal '\0' */
537 if (strnlen(filename.name, filename.len) != filename.len)
540 if (strnchr(filename.name, filename.len, '/'))
542 if (filename.name[0] == '.') {
543 if (filename.len == 1)
545 if (filename.len == 2 && filename.name[1] == '.')
548 filename.hash = full_name_hash(parent, filename.name, filename.len);
550 dentry = d_lookup(parent, &filename);
553 dentry = d_alloc_parallel(parent, &filename, &wq);
557 if (!d_in_lookup(dentry)) {
558 /* Is there a mountpoint here? If so, just exit */
559 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
560 &entry->fattr->fsid))
562 if (nfs_same_file(dentry, entry)) {
563 if (!entry->fh->size)
565 nfs_set_verifier(dentry, dir_verifier);
566 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
568 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
571 d_invalidate(dentry);
577 if (!entry->fh->size) {
578 d_lookup_done(dentry);
582 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
583 alias = d_splice_alias(inode, dentry);
584 d_lookup_done(dentry);
591 nfs_set_verifier(dentry, dir_verifier);
596 /* Perform conversion from xdr to cache array */
598 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
599 struct page **xdr_pages, struct page *page, unsigned int buflen)
601 struct xdr_stream stream;
603 struct page *scratch;
606 scratch = alloc_page(GFP_KERNEL);
610 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
611 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
615 entry->label->len = NFS4_MAXLABELLEN;
617 status = xdr_decode(desc, entry, &stream);
622 nfs_prime_dcache(file_dentry(desc->file), entry,
625 status = nfs_readdir_add_to_array(entry, page);
626 } while (!status && !entry->eof);
631 nfs_readdir_page_set_eof(page);
646 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
649 for (i = 0; i < npages; i++)
654 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
655 * to nfs_readdir_free_pages()
658 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
662 for (i = 0; i < npages; i++) {
663 struct page *page = alloc_page(GFP_KERNEL);
671 nfs_readdir_free_pages(pages, i);
676 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
678 struct page *pages[NFS_MAX_READDIR_PAGES];
679 struct nfs_entry entry;
680 struct file *file = desc->file;
681 struct nfs_cache_array *array;
682 int status = -ENOMEM;
683 unsigned int array_size = ARRAY_SIZE(pages);
685 nfs_readdir_init_array(page);
687 entry.prev_cookie = 0;
688 entry.cookie = desc->last_cookie;
690 entry.fh = nfs_alloc_fhandle();
691 entry.fattr = nfs_alloc_fattr();
692 entry.server = NFS_SERVER(inode);
693 if (entry.fh == NULL || entry.fattr == NULL)
696 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
697 if (IS_ERR(entry.label)) {
698 status = PTR_ERR(entry.label);
704 status = nfs_readdir_alloc_pages(pages, array_size);
706 goto out_release_array;
709 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
716 nfs_readdir_page_set_eof(page);
720 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
721 } while (!status && !nfs_readdir_array_is_full(array));
723 nfs_readdir_free_pages(pages, array_size);
726 nfs4_label_free(entry.label);
728 nfs_free_fattr(entry.fattr);
729 nfs_free_fhandle(entry.fh);
734 * Now we cache directories properly, by converting xdr information
735 * to an array that can be used for lookups later. This results in
736 * fewer cache pages, since we can store more information on each page.
737 * We only need to convert from xdr once so future lookups are much simpler
740 int nfs_readdir_filler(void *data, struct page* page)
742 nfs_readdir_descriptor_t *desc = data;
743 struct inode *inode = file_inode(desc->file);
746 ret = nfs_readdir_xdr_to_array(desc, page, inode);
749 SetPageUptodate(page);
751 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
752 /* Should never happen */
753 nfs_zap_mapping(inode, inode->i_mapping);
758 nfs_readdir_clear_array(page);
764 void cache_page_release(nfs_readdir_descriptor_t *desc)
766 put_page(desc->page);
771 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
773 return read_cache_page(desc->file->f_mapping, desc->page_index,
774 nfs_readdir_filler, desc);
778 * Returns 0 if desc->dir_cookie was found on page desc->page_index
779 * and locks the page to prevent removal from the page cache.
782 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
784 struct inode *inode = file_inode(desc->file);
785 struct nfs_inode *nfsi = NFS_I(inode);
788 desc->page = get_cache_page(desc);
789 if (IS_ERR(desc->page))
790 return PTR_ERR(desc->page);
791 res = lock_page_killable(desc->page);
795 if (desc->page->mapping != NULL) {
796 res = nfs_readdir_search_array(desc);
798 nfsi->page_index = desc->page_index;
802 unlock_page(desc->page);
804 cache_page_release(desc);
808 /* Search for desc->dir_cookie from the beginning of the page cache */
810 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
814 if (desc->page_index == 0) {
815 desc->current_index = 0;
816 desc->prev_index = 0;
817 desc->last_cookie = 0;
820 res = find_and_lock_cache_page(desc);
821 } while (res == -EAGAIN);
826 * Once we've found the start of the dirent within a page: fill 'er up...
829 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
831 struct file *file = desc->file;
834 struct nfs_cache_array *array = NULL;
836 array = kmap(desc->page);
837 for (i = desc->cache_entry_index; i < array->size; i++) {
838 struct nfs_cache_array_entry *ent;
840 ent = &array->array[i];
841 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
842 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
846 if (i < (array->size-1))
847 desc->dir_cookie = array->array[i+1].cookie;
849 desc->dir_cookie = array->last_cookie;
850 if (nfs_readdir_use_cookie(file))
851 desc->ctx->pos = desc->dir_cookie;
854 if (desc->duped != 0)
857 if (array->page_is_eof)
861 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
862 (unsigned long long)desc->dir_cookie, res);
867 * If we cannot find a cookie in our cache, we suspect that this is
868 * because it points to a deleted file, so we ask the server to return
869 * whatever it thinks is the next entry. We then feed this to filldir.
870 * If all goes well, we should then be able to find our way round the
871 * cache on the next call to readdir_search_pagecache();
873 * NOTE: we cannot add the anonymous page to the pagecache because
874 * the data it contains might not be page aligned. Besides,
875 * we should already have a complete representation of the
876 * directory in the page cache by the time we get here.
879 int uncached_readdir(nfs_readdir_descriptor_t *desc)
881 struct page *page = NULL;
883 struct inode *inode = file_inode(desc->file);
885 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
886 (unsigned long long)desc->dir_cookie);
888 page = alloc_page(GFP_HIGHUSER);
894 desc->page_index = 0;
895 desc->last_cookie = desc->dir_cookie;
899 status = nfs_readdir_xdr_to_array(desc, page, inode);
903 status = nfs_do_filldir(desc);
906 nfs_readdir_clear_array(desc->page);
907 cache_page_release(desc);
909 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
914 /* The file offset position represents the dirent entry number. A
915 last cookie cache takes care of the common case of reading the
918 static int nfs_readdir(struct file *file, struct dir_context *ctx)
920 struct dentry *dentry = file_dentry(file);
921 struct inode *inode = d_inode(dentry);
922 struct nfs_open_dir_context *dir_ctx = file->private_data;
923 nfs_readdir_descriptor_t my_desc = {
926 .plus = nfs_use_readdirplus(inode, ctx),
931 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
932 file, (long long)ctx->pos);
933 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
936 * ctx->pos points to the dirent entry number.
937 * *desc->dir_cookie has the cookie for the next entry. We have
938 * to either find the entry with the appropriate number or
939 * revalidate the cookie.
941 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
942 res = nfs_revalidate_mapping(inode, file->f_mapping);
946 spin_lock(&file->f_lock);
947 desc->dir_cookie = dir_ctx->dir_cookie;
948 desc->dup_cookie = dir_ctx->dup_cookie;
949 desc->duped = dir_ctx->duped;
950 desc->attr_gencount = dir_ctx->attr_gencount;
951 spin_unlock(&file->f_lock);
954 res = readdir_search_pagecache(desc);
956 if (res == -EBADCOOKIE) {
958 /* This means either end of directory */
959 if (desc->dir_cookie && !desc->eof) {
960 /* Or that the server has 'lost' a cookie */
961 res = uncached_readdir(desc);
967 if (res == -ETOOSMALL && desc->plus) {
968 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
969 nfs_zap_caches(inode);
970 desc->page_index = 0;
978 res = nfs_do_filldir(desc);
979 unlock_page(desc->page);
980 cache_page_release(desc);
983 } while (!desc->eof);
985 spin_lock(&file->f_lock);
986 dir_ctx->dir_cookie = desc->dir_cookie;
987 dir_ctx->dup_cookie = desc->dup_cookie;
988 dir_ctx->duped = desc->duped;
989 dir_ctx->attr_gencount = desc->attr_gencount;
990 spin_unlock(&file->f_lock);
995 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
999 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1001 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1003 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1004 filp, offset, whence);
1012 spin_lock(&filp->f_lock);
1017 spin_lock(&filp->f_lock);
1018 offset += filp->f_pos;
1020 spin_unlock(&filp->f_lock);
1024 if (offset != filp->f_pos) {
1025 filp->f_pos = offset;
1026 if (nfs_readdir_use_cookie(filp))
1027 dir_ctx->dir_cookie = offset;
1029 dir_ctx->dir_cookie = 0;
1032 spin_unlock(&filp->f_lock);
1037 * All directory operations under NFS are synchronous, so fsync()
1038 * is a dummy operation.
1040 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1043 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1045 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1050 * nfs_force_lookup_revalidate - Mark the directory as having changed
1051 * @dir: pointer to directory inode
1053 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1054 * full lookup on all child dentries of 'dir' whenever a change occurs
1055 * on the server that might have invalidated our dcache.
1057 * Note that we reserve bit '0' as a tag to let us know when a dentry
1058 * was revalidated while holding a delegation on its inode.
1060 * The caller should be holding dir->i_lock
1062 void nfs_force_lookup_revalidate(struct inode *dir)
1064 NFS_I(dir)->cache_change_attribute += 2;
1066 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1069 * nfs_verify_change_attribute - Detects NFS remote directory changes
1070 * @dir: pointer to parent directory inode
1071 * @verf: previously saved change attribute
1073 * Return "false" if the verifiers doesn't match the change attribute.
1074 * This would usually indicate that the directory contents have changed on
1075 * the server, and that any dentries need revalidating.
1077 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1079 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1082 static void nfs_set_verifier_delegated(unsigned long *verf)
1087 #if IS_ENABLED(CONFIG_NFS_V4)
1088 static void nfs_unset_verifier_delegated(unsigned long *verf)
1092 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1094 static bool nfs_test_verifier_delegated(unsigned long verf)
1099 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1101 return nfs_test_verifier_delegated(dentry->d_time);
1104 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1106 struct inode *inode = d_inode(dentry);
1108 if (!nfs_verifier_is_delegated(dentry) &&
1109 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1111 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1112 nfs_set_verifier_delegated(&verf);
1114 dentry->d_time = verf;
1118 * nfs_set_verifier - save a parent directory verifier in the dentry
1119 * @dentry: pointer to dentry
1120 * @verf: verifier to save
1122 * Saves the parent directory verifier in @dentry. If the inode has
1123 * a delegation, we also tag the dentry as having been revalidated
1124 * while holding a delegation so that we know we don't have to
1125 * look it up again after a directory change.
1127 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1130 spin_lock(&dentry->d_lock);
1131 nfs_set_verifier_locked(dentry, verf);
1132 spin_unlock(&dentry->d_lock);
1134 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1136 #if IS_ENABLED(CONFIG_NFS_V4)
1138 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1139 * @inode: pointer to inode
1141 * Iterates through the dentries in the inode alias list and clears
1142 * the tag used to indicate that the dentry has been revalidated
1143 * while holding a delegation.
1144 * This function is intended for use when the delegation is being
1145 * returned or revoked.
1147 void nfs_clear_verifier_delegated(struct inode *inode)
1149 struct dentry *alias;
1153 spin_lock(&inode->i_lock);
1154 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1155 spin_lock(&alias->d_lock);
1156 nfs_unset_verifier_delegated(&alias->d_time);
1157 spin_unlock(&alias->d_lock);
1159 spin_unlock(&inode->i_lock);
1161 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1162 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1165 * A check for whether or not the parent directory has changed.
1166 * In the case it has, we assume that the dentries are untrustworthy
1167 * and may need to be looked up again.
1168 * If rcu_walk prevents us from performing a full check, return 0.
1170 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1173 if (IS_ROOT(dentry))
1175 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1177 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1179 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1180 if (nfs_mapping_need_revalidate_inode(dir)) {
1183 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1186 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1192 * Use intent information to check whether or not we're going to do
1193 * an O_EXCL create using this path component.
1195 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1197 if (NFS_PROTO(dir)->version == 2)
1199 return flags & LOOKUP_EXCL;
1203 * Inode and filehandle revalidation for lookups.
1205 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1206 * or if the intent information indicates that we're about to open this
1207 * particular file and the "nocto" mount flag is not set.
1211 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1213 struct nfs_server *server = NFS_SERVER(inode);
1216 if (IS_AUTOMOUNT(inode))
1219 if (flags & LOOKUP_OPEN) {
1220 switch (inode->i_mode & S_IFMT) {
1222 /* A NFSv4 OPEN will revalidate later */
1223 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1227 if (server->flags & NFS_MOUNT_NOCTO)
1229 /* NFS close-to-open cache consistency validation */
1234 /* VFS wants an on-the-wire revalidation */
1235 if (flags & LOOKUP_REVAL)
1238 return (inode->i_nlink == 0) ? -ESTALE : 0;
1240 if (flags & LOOKUP_RCU)
1242 ret = __nfs_revalidate_inode(server, inode);
1249 * We judge how long we want to trust negative
1250 * dentries by looking at the parent inode mtime.
1252 * If parent mtime has changed, we revalidate, else we wait for a
1253 * period corresponding to the parent's attribute cache timeout value.
1255 * If LOOKUP_RCU prevents us from performing a full check, return 1
1256 * suggesting a reval is needed.
1258 * Note that when creating a new file, or looking up a rename target,
1259 * then it shouldn't be necessary to revalidate a negative dentry.
1262 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1265 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1267 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1269 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1273 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1274 struct inode *inode, int error)
1278 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1282 nfs_mark_for_revalidate(dir);
1283 if (inode && S_ISDIR(inode->i_mode)) {
1284 /* Purge readdir caches. */
1285 nfs_zap_caches(inode);
1287 * We can't d_drop the root of a disconnected tree:
1288 * its d_hash is on the s_anon list and d_drop() would hide
1289 * it from shrink_dcache_for_unmount(), leading to busy
1290 * inodes on unmount and further oopses.
1292 if (IS_ROOT(dentry))
1295 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1299 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1300 __func__, dentry, error);
1305 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1309 if (nfs_neg_need_reval(dir, dentry, flags)) {
1310 if (flags & LOOKUP_RCU)
1314 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1318 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1319 struct inode *inode)
1321 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1322 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1326 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1327 struct inode *inode)
1329 struct nfs_fh *fhandle;
1330 struct nfs_fattr *fattr;
1331 struct nfs4_label *label;
1332 unsigned long dir_verifier;
1336 fhandle = nfs_alloc_fhandle();
1337 fattr = nfs_alloc_fattr();
1338 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1339 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1342 dir_verifier = nfs_save_change_attribute(dir);
1343 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1351 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1357 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1359 if (nfs_refresh_inode(inode, fattr) < 0)
1362 nfs_setsecurity(inode, fattr, label);
1363 nfs_set_verifier(dentry, dir_verifier);
1365 /* set a readdirplus hint that we had a cache miss */
1366 nfs_force_use_readdirplus(dir);
1369 nfs_free_fattr(fattr);
1370 nfs_free_fhandle(fhandle);
1371 nfs4_label_free(label);
1372 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1376 * This is called every time the dcache has a lookup hit,
1377 * and we should check whether we can really trust that
1380 * NOTE! The hit can be a negative hit too, don't assume
1383 * If the parent directory is seen to have changed, we throw out the
1384 * cached dentry and do a new lookup.
1387 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1390 struct inode *inode;
1393 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1394 inode = d_inode(dentry);
1397 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1399 if (is_bad_inode(inode)) {
1400 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1405 if (nfs_verifier_is_delegated(dentry))
1406 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1408 /* Force a full look up iff the parent directory has changed */
1409 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1410 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1411 error = nfs_lookup_verify_inode(inode, flags);
1413 if (error == -ESTALE)
1414 nfs_zap_caches(dir);
1417 nfs_advise_use_readdirplus(dir);
1421 if (flags & LOOKUP_RCU)
1424 if (NFS_STALE(inode))
1427 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1428 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1429 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1432 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1434 if (flags & LOOKUP_RCU)
1436 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1440 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1441 int (*reval)(struct inode *, struct dentry *, unsigned int))
1443 struct dentry *parent;
1447 if (flags & LOOKUP_RCU) {
1448 parent = READ_ONCE(dentry->d_parent);
1449 dir = d_inode_rcu(parent);
1452 ret = reval(dir, dentry, flags);
1453 if (parent != READ_ONCE(dentry->d_parent))
1456 parent = dget_parent(dentry);
1457 ret = reval(d_inode(parent), dentry, flags);
1463 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1465 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1469 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1470 * when we don't really care about the dentry name. This is called when a
1471 * pathwalk ends on a dentry that was not found via a normal lookup in the
1472 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1474 * In this situation, we just want to verify that the inode itself is OK
1475 * since the dentry might have changed on the server.
1477 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1479 struct inode *inode = d_inode(dentry);
1483 * I believe we can only get a negative dentry here in the case of a
1484 * procfs-style symlink. Just assume it's correct for now, but we may
1485 * eventually need to do something more here.
1488 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1493 if (is_bad_inode(inode)) {
1494 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1499 error = nfs_lookup_verify_inode(inode, flags);
1500 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1501 __func__, inode->i_ino, error ? "invalid" : "valid");
1506 * This is called from dput() when d_count is going to 0.
1508 static int nfs_dentry_delete(const struct dentry *dentry)
1510 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1511 dentry, dentry->d_flags);
1513 /* Unhash any dentry with a stale inode */
1514 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1517 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1518 /* Unhash it, so that ->d_iput() would be called */
1521 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1522 /* Unhash it, so that ancestors of killed async unlink
1523 * files will be cleaned up during umount */
1530 /* Ensure that we revalidate inode->i_nlink */
1531 static void nfs_drop_nlink(struct inode *inode)
1533 spin_lock(&inode->i_lock);
1534 /* drop the inode if we're reasonably sure this is the last link */
1535 if (inode->i_nlink > 0)
1537 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1538 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1539 | NFS_INO_INVALID_CTIME
1540 | NFS_INO_INVALID_OTHER
1541 | NFS_INO_REVAL_FORCED;
1542 spin_unlock(&inode->i_lock);
1546 * Called when the dentry loses inode.
1547 * We use it to clean up silly-renamed files.
1549 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1551 if (S_ISDIR(inode->i_mode))
1552 /* drop any readdir cache as it could easily be old */
1553 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1555 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1556 nfs_complete_unlink(dentry, inode);
1557 nfs_drop_nlink(inode);
1562 static void nfs_d_release(struct dentry *dentry)
1564 /* free cached devname value, if it survived that far */
1565 if (unlikely(dentry->d_fsdata)) {
1566 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1569 kfree(dentry->d_fsdata);
1573 const struct dentry_operations nfs_dentry_operations = {
1574 .d_revalidate = nfs_lookup_revalidate,
1575 .d_weak_revalidate = nfs_weak_revalidate,
1576 .d_delete = nfs_dentry_delete,
1577 .d_iput = nfs_dentry_iput,
1578 .d_automount = nfs_d_automount,
1579 .d_release = nfs_d_release,
1581 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1583 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1586 struct inode *inode = NULL;
1587 struct nfs_fh *fhandle = NULL;
1588 struct nfs_fattr *fattr = NULL;
1589 struct nfs4_label *label = NULL;
1590 unsigned long dir_verifier;
1593 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1594 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1596 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1597 return ERR_PTR(-ENAMETOOLONG);
1600 * If we're doing an exclusive create, optimize away the lookup
1601 * but don't hash the dentry.
1603 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1606 res = ERR_PTR(-ENOMEM);
1607 fhandle = nfs_alloc_fhandle();
1608 fattr = nfs_alloc_fattr();
1609 if (fhandle == NULL || fattr == NULL)
1612 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1616 dir_verifier = nfs_save_change_attribute(dir);
1617 trace_nfs_lookup_enter(dir, dentry, flags);
1618 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1619 if (error == -ENOENT)
1622 res = ERR_PTR(error);
1625 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1626 res = ERR_CAST(inode);
1630 /* Notify readdir to use READDIRPLUS */
1631 nfs_force_use_readdirplus(dir);
1634 res = d_splice_alias(inode, dentry);
1640 nfs_set_verifier(dentry, dir_verifier);
1642 trace_nfs_lookup_exit(dir, dentry, flags, error);
1643 nfs4_label_free(label);
1645 nfs_free_fattr(fattr);
1646 nfs_free_fhandle(fhandle);
1649 EXPORT_SYMBOL_GPL(nfs_lookup);
1651 #if IS_ENABLED(CONFIG_NFS_V4)
1652 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1654 const struct dentry_operations nfs4_dentry_operations = {
1655 .d_revalidate = nfs4_lookup_revalidate,
1656 .d_weak_revalidate = nfs_weak_revalidate,
1657 .d_delete = nfs_dentry_delete,
1658 .d_iput = nfs_dentry_iput,
1659 .d_automount = nfs_d_automount,
1660 .d_release = nfs_d_release,
1662 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1664 static fmode_t flags_to_mode(int flags)
1666 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1667 if ((flags & O_ACCMODE) != O_WRONLY)
1669 if ((flags & O_ACCMODE) != O_RDONLY)
1674 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1676 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1679 static int do_open(struct inode *inode, struct file *filp)
1681 nfs_fscache_open_file(inode, filp);
1685 static int nfs_finish_open(struct nfs_open_context *ctx,
1686 struct dentry *dentry,
1687 struct file *file, unsigned open_flags)
1691 err = finish_open(file, dentry, do_open);
1694 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1695 nfs_file_set_open_context(file, ctx);
1702 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1703 struct file *file, unsigned open_flags,
1706 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1707 struct nfs_open_context *ctx;
1709 struct iattr attr = { .ia_valid = ATTR_OPEN };
1710 struct inode *inode;
1711 unsigned int lookup_flags = 0;
1712 bool switched = false;
1716 /* Expect a negative dentry */
1717 BUG_ON(d_inode(dentry));
1719 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1720 dir->i_sb->s_id, dir->i_ino, dentry);
1722 err = nfs_check_flags(open_flags);
1726 /* NFS only supports OPEN on regular files */
1727 if ((open_flags & O_DIRECTORY)) {
1728 if (!d_in_lookup(dentry)) {
1730 * Hashed negative dentry with O_DIRECTORY: dentry was
1731 * revalidated and is fine, no need to perform lookup
1736 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1740 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1741 return -ENAMETOOLONG;
1743 if (open_flags & O_CREAT) {
1744 struct nfs_server *server = NFS_SERVER(dir);
1746 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1747 mode &= ~current_umask();
1749 attr.ia_valid |= ATTR_MODE;
1750 attr.ia_mode = mode;
1752 if (open_flags & O_TRUNC) {
1753 attr.ia_valid |= ATTR_SIZE;
1757 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1760 dentry = d_alloc_parallel(dentry->d_parent,
1761 &dentry->d_name, &wq);
1763 return PTR_ERR(dentry);
1764 if (unlikely(!d_in_lookup(dentry)))
1765 return finish_no_open(file, dentry);
1768 ctx = create_nfs_open_context(dentry, open_flags, file);
1773 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1774 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1776 file->f_mode |= FMODE_CREATED;
1777 if (IS_ERR(inode)) {
1778 err = PTR_ERR(inode);
1779 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1780 put_nfs_open_context(ctx);
1784 d_splice_alias(NULL, dentry);
1785 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1791 if (!(open_flags & O_NOFOLLOW))
1801 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1802 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1803 put_nfs_open_context(ctx);
1805 if (unlikely(switched)) {
1806 d_lookup_done(dentry);
1812 res = nfs_lookup(dir, dentry, lookup_flags);
1814 d_lookup_done(dentry);
1821 return PTR_ERR(res);
1822 return finish_no_open(file, res);
1824 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1827 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1830 struct inode *inode;
1832 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1834 if (d_mountpoint(dentry))
1837 inode = d_inode(dentry);
1839 /* We can't create new files in nfs_open_revalidate(), so we
1840 * optimize away revalidation of negative dentries.
1845 if (nfs_verifier_is_delegated(dentry))
1846 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1848 /* NFS only supports OPEN on regular files */
1849 if (!S_ISREG(inode->i_mode))
1852 /* We cannot do exclusive creation on a positive dentry */
1853 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1856 /* Check if the directory changed */
1857 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1860 /* Let f_op->open() actually open (and revalidate) the file */
1863 if (flags & LOOKUP_RCU)
1865 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1868 return nfs_do_lookup_revalidate(dir, dentry, flags);
1871 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1873 return __nfs_lookup_revalidate(dentry, flags,
1874 nfs4_do_lookup_revalidate);
1877 #endif /* CONFIG_NFSV4 */
1880 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1881 struct nfs_fattr *fattr,
1882 struct nfs4_label *label)
1884 struct dentry *parent = dget_parent(dentry);
1885 struct inode *dir = d_inode(parent);
1886 struct inode *inode;
1892 if (fhandle->size == 0) {
1893 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
1897 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1898 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1899 struct nfs_server *server = NFS_SB(dentry->d_sb);
1900 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1905 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1906 d = d_splice_alias(inode, dentry);
1911 nfs_mark_for_revalidate(dir);
1915 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1918 * Code common to create, mkdir, and mknod.
1920 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1921 struct nfs_fattr *fattr,
1922 struct nfs4_label *label)
1926 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1930 /* Callers don't care */
1934 EXPORT_SYMBOL_GPL(nfs_instantiate);
1937 * Following a failed create operation, we drop the dentry rather
1938 * than retain a negative dentry. This avoids a problem in the event
1939 * that the operation succeeded on the server, but an error in the
1940 * reply path made it appear to have failed.
1942 int nfs_create(struct inode *dir, struct dentry *dentry,
1943 umode_t mode, bool excl)
1946 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1949 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1950 dir->i_sb->s_id, dir->i_ino, dentry);
1952 attr.ia_mode = mode;
1953 attr.ia_valid = ATTR_MODE;
1955 trace_nfs_create_enter(dir, dentry, open_flags);
1956 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1957 trace_nfs_create_exit(dir, dentry, open_flags, error);
1965 EXPORT_SYMBOL_GPL(nfs_create);
1968 * See comments for nfs_proc_create regarding failed operations.
1971 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1976 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1977 dir->i_sb->s_id, dir->i_ino, dentry);
1979 attr.ia_mode = mode;
1980 attr.ia_valid = ATTR_MODE;
1982 trace_nfs_mknod_enter(dir, dentry);
1983 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1984 trace_nfs_mknod_exit(dir, dentry, status);
1992 EXPORT_SYMBOL_GPL(nfs_mknod);
1995 * See comments for nfs_proc_create regarding failed operations.
1997 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2002 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2003 dir->i_sb->s_id, dir->i_ino, dentry);
2005 attr.ia_valid = ATTR_MODE;
2006 attr.ia_mode = mode | S_IFDIR;
2008 trace_nfs_mkdir_enter(dir, dentry);
2009 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2010 trace_nfs_mkdir_exit(dir, dentry, error);
2018 EXPORT_SYMBOL_GPL(nfs_mkdir);
2020 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2022 if (simple_positive(dentry))
2026 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2030 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2031 dir->i_sb->s_id, dir->i_ino, dentry);
2033 trace_nfs_rmdir_enter(dir, dentry);
2034 if (d_really_is_positive(dentry)) {
2035 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2036 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2037 /* Ensure the VFS deletes this inode */
2040 clear_nlink(d_inode(dentry));
2043 nfs_dentry_handle_enoent(dentry);
2045 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2047 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2048 trace_nfs_rmdir_exit(dir, dentry, error);
2052 EXPORT_SYMBOL_GPL(nfs_rmdir);
2055 * Remove a file after making sure there are no pending writes,
2056 * and after checking that the file has only one user.
2058 * We invalidate the attribute cache and free the inode prior to the operation
2059 * to avoid possible races if the server reuses the inode.
2061 static int nfs_safe_remove(struct dentry *dentry)
2063 struct inode *dir = d_inode(dentry->d_parent);
2064 struct inode *inode = d_inode(dentry);
2067 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2069 /* If the dentry was sillyrenamed, we simply call d_delete() */
2070 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2075 trace_nfs_remove_enter(dir, dentry);
2076 if (inode != NULL) {
2077 error = NFS_PROTO(dir)->remove(dir, dentry);
2079 nfs_drop_nlink(inode);
2081 error = NFS_PROTO(dir)->remove(dir, dentry);
2082 if (error == -ENOENT)
2083 nfs_dentry_handle_enoent(dentry);
2084 trace_nfs_remove_exit(dir, dentry, error);
2089 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2090 * belongs to an active ".nfs..." file and we return -EBUSY.
2092 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2094 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2097 int need_rehash = 0;
2099 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2100 dir->i_ino, dentry);
2102 trace_nfs_unlink_enter(dir, dentry);
2103 spin_lock(&dentry->d_lock);
2104 if (d_count(dentry) > 1) {
2105 spin_unlock(&dentry->d_lock);
2106 /* Start asynchronous writeout of the inode */
2107 write_inode_now(d_inode(dentry), 0);
2108 error = nfs_sillyrename(dir, dentry);
2111 if (!d_unhashed(dentry)) {
2115 spin_unlock(&dentry->d_lock);
2116 error = nfs_safe_remove(dentry);
2117 if (!error || error == -ENOENT) {
2118 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2119 } else if (need_rehash)
2122 trace_nfs_unlink_exit(dir, dentry, error);
2125 EXPORT_SYMBOL_GPL(nfs_unlink);
2128 * To create a symbolic link, most file systems instantiate a new inode,
2129 * add a page to it containing the path, then write it out to the disk
2130 * using prepare_write/commit_write.
2132 * Unfortunately the NFS client can't create the in-core inode first
2133 * because it needs a file handle to create an in-core inode (see
2134 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2135 * symlink request has completed on the server.
2137 * So instead we allocate a raw page, copy the symname into it, then do
2138 * the SYMLINK request with the page as the buffer. If it succeeds, we
2139 * now have a new file handle and can instantiate an in-core NFS inode
2140 * and move the raw page into its mapping.
2142 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2147 unsigned int pathlen = strlen(symname);
2150 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2151 dir->i_ino, dentry, symname);
2153 if (pathlen > PAGE_SIZE)
2154 return -ENAMETOOLONG;
2156 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2157 attr.ia_valid = ATTR_MODE;
2159 page = alloc_page(GFP_USER);
2163 kaddr = page_address(page);
2164 memcpy(kaddr, symname, pathlen);
2165 if (pathlen < PAGE_SIZE)
2166 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2168 trace_nfs_symlink_enter(dir, dentry);
2169 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2170 trace_nfs_symlink_exit(dir, dentry, error);
2172 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2173 dir->i_sb->s_id, dir->i_ino,
2174 dentry, symname, error);
2181 * No big deal if we can't add this page to the page cache here.
2182 * READLINK will get the missing page from the server if needed.
2184 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2186 SetPageUptodate(page);
2189 * add_to_page_cache_lru() grabs an extra page refcount.
2190 * Drop it here to avoid leaking this page later.
2198 EXPORT_SYMBOL_GPL(nfs_symlink);
2201 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2203 struct inode *inode = d_inode(old_dentry);
2206 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2207 old_dentry, dentry);
2209 trace_nfs_link_enter(inode, dir, dentry);
2211 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2214 d_add(dentry, inode);
2216 trace_nfs_link_exit(inode, dir, dentry, error);
2219 EXPORT_SYMBOL_GPL(nfs_link);
2223 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2224 * different file handle for the same inode after a rename (e.g. when
2225 * moving to a different directory). A fail-safe method to do so would
2226 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2227 * rename the old file using the sillyrename stuff. This way, the original
2228 * file in old_dir will go away when the last process iput()s the inode.
2232 * It actually works quite well. One needs to have the possibility for
2233 * at least one ".nfs..." file in each directory the file ever gets
2234 * moved or linked to which happens automagically with the new
2235 * implementation that only depends on the dcache stuff instead of
2236 * using the inode layer
2238 * Unfortunately, things are a little more complicated than indicated
2239 * above. For a cross-directory move, we want to make sure we can get
2240 * rid of the old inode after the operation. This means there must be
2241 * no pending writes (if it's a file), and the use count must be 1.
2242 * If these conditions are met, we can drop the dentries before doing
2245 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2246 struct inode *new_dir, struct dentry *new_dentry,
2249 struct inode *old_inode = d_inode(old_dentry);
2250 struct inode *new_inode = d_inode(new_dentry);
2251 struct dentry *dentry = NULL, *rehash = NULL;
2252 struct rpc_task *task;
2258 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2259 old_dentry, new_dentry,
2260 d_count(new_dentry));
2262 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2264 * For non-directories, check whether the target is busy and if so,
2265 * make a copy of the dentry and then do a silly-rename. If the
2266 * silly-rename succeeds, the copied dentry is hashed and becomes
2269 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2271 * To prevent any new references to the target during the
2272 * rename, we unhash the dentry in advance.
2274 if (!d_unhashed(new_dentry)) {
2276 rehash = new_dentry;
2279 if (d_count(new_dentry) > 2) {
2282 /* copy the target dentry's name */
2283 dentry = d_alloc(new_dentry->d_parent,
2284 &new_dentry->d_name);
2288 /* silly-rename the existing target ... */
2289 err = nfs_sillyrename(new_dir, new_dentry);
2293 new_dentry = dentry;
2299 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2301 error = PTR_ERR(task);
2305 error = rpc_wait_for_completion_task(task);
2307 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2308 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2311 error = task->tk_status;
2313 /* Ensure the inode attributes are revalidated */
2315 spin_lock(&old_inode->i_lock);
2316 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2317 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2318 | NFS_INO_INVALID_CTIME
2319 | NFS_INO_REVAL_FORCED;
2320 spin_unlock(&old_inode->i_lock);
2325 trace_nfs_rename_exit(old_dir, old_dentry,
2326 new_dir, new_dentry, error);
2328 if (new_inode != NULL)
2329 nfs_drop_nlink(new_inode);
2331 * The d_move() should be here instead of in an async RPC completion
2332 * handler because we need the proper locks to move the dentry. If
2333 * we're interrupted by a signal, the async RPC completion handler
2334 * should mark the directories for revalidation.
2336 d_move(old_dentry, new_dentry);
2337 nfs_set_verifier(old_dentry,
2338 nfs_save_change_attribute(new_dir));
2339 } else if (error == -ENOENT)
2340 nfs_dentry_handle_enoent(old_dentry);
2342 /* new dentry created? */
2347 EXPORT_SYMBOL_GPL(nfs_rename);
2349 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2350 static LIST_HEAD(nfs_access_lru_list);
2351 static atomic_long_t nfs_access_nr_entries;
2353 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2354 module_param(nfs_access_max_cachesize, ulong, 0644);
2355 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2357 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2359 put_cred(entry->cred);
2360 kfree_rcu(entry, rcu_head);
2361 smp_mb__before_atomic();
2362 atomic_long_dec(&nfs_access_nr_entries);
2363 smp_mb__after_atomic();
2366 static void nfs_access_free_list(struct list_head *head)
2368 struct nfs_access_entry *cache;
2370 while (!list_empty(head)) {
2371 cache = list_entry(head->next, struct nfs_access_entry, lru);
2372 list_del(&cache->lru);
2373 nfs_access_free_entry(cache);
2377 static unsigned long
2378 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2381 struct nfs_inode *nfsi, *next;
2382 struct nfs_access_entry *cache;
2385 spin_lock(&nfs_access_lru_lock);
2386 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2387 struct inode *inode;
2389 if (nr_to_scan-- == 0)
2391 inode = &nfsi->vfs_inode;
2392 spin_lock(&inode->i_lock);
2393 if (list_empty(&nfsi->access_cache_entry_lru))
2394 goto remove_lru_entry;
2395 cache = list_entry(nfsi->access_cache_entry_lru.next,
2396 struct nfs_access_entry, lru);
2397 list_move(&cache->lru, &head);
2398 rb_erase(&cache->rb_node, &nfsi->access_cache);
2400 if (!list_empty(&nfsi->access_cache_entry_lru))
2401 list_move_tail(&nfsi->access_cache_inode_lru,
2402 &nfs_access_lru_list);
2405 list_del_init(&nfsi->access_cache_inode_lru);
2406 smp_mb__before_atomic();
2407 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2408 smp_mb__after_atomic();
2410 spin_unlock(&inode->i_lock);
2412 spin_unlock(&nfs_access_lru_lock);
2413 nfs_access_free_list(&head);
2418 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2420 int nr_to_scan = sc->nr_to_scan;
2421 gfp_t gfp_mask = sc->gfp_mask;
2423 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2425 return nfs_do_access_cache_scan(nr_to_scan);
2430 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2432 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2436 nfs_access_cache_enforce_limit(void)
2438 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2440 unsigned int nr_to_scan;
2442 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2445 diff = nr_entries - nfs_access_max_cachesize;
2446 if (diff < nr_to_scan)
2448 nfs_do_access_cache_scan(nr_to_scan);
2451 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2453 struct rb_root *root_node = &nfsi->access_cache;
2455 struct nfs_access_entry *entry;
2457 /* Unhook entries from the cache */
2458 while ((n = rb_first(root_node)) != NULL) {
2459 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2460 rb_erase(n, root_node);
2461 list_move(&entry->lru, head);
2463 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2466 void nfs_access_zap_cache(struct inode *inode)
2470 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2472 /* Remove from global LRU init */
2473 spin_lock(&nfs_access_lru_lock);
2474 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2475 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2477 spin_lock(&inode->i_lock);
2478 __nfs_access_zap_cache(NFS_I(inode), &head);
2479 spin_unlock(&inode->i_lock);
2480 spin_unlock(&nfs_access_lru_lock);
2481 nfs_access_free_list(&head);
2483 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2485 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2487 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2490 struct nfs_access_entry *entry =
2491 rb_entry(n, struct nfs_access_entry, rb_node);
2492 int cmp = cred_fscmp(cred, entry->cred);
2504 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2506 struct nfs_inode *nfsi = NFS_I(inode);
2507 struct nfs_access_entry *cache;
2511 spin_lock(&inode->i_lock);
2513 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2515 cache = nfs_access_search_rbtree(inode, cred);
2519 /* Found an entry, is our attribute cache valid? */
2520 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2527 spin_unlock(&inode->i_lock);
2528 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2531 spin_lock(&inode->i_lock);
2534 res->cred = cache->cred;
2535 res->mask = cache->mask;
2536 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2539 spin_unlock(&inode->i_lock);
2542 spin_unlock(&inode->i_lock);
2543 nfs_access_zap_cache(inode);
2547 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2549 /* Only check the most recently returned cache entry,
2550 * but do it without locking.
2552 struct nfs_inode *nfsi = NFS_I(inode);
2553 struct nfs_access_entry *cache;
2555 struct list_head *lh;
2558 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2560 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2561 cache = list_entry(lh, struct nfs_access_entry, lru);
2562 if (lh == &nfsi->access_cache_entry_lru ||
2563 cred_fscmp(cred, cache->cred) != 0)
2567 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2569 res->cred = cache->cred;
2570 res->mask = cache->mask;
2577 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2578 nfs_access_entry *res, bool may_block)
2582 status = nfs_access_get_cached_rcu(inode, cred, res);
2584 status = nfs_access_get_cached_locked(inode, cred, res,
2589 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2591 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2593 struct nfs_inode *nfsi = NFS_I(inode);
2594 struct rb_root *root_node = &nfsi->access_cache;
2595 struct rb_node **p = &root_node->rb_node;
2596 struct rb_node *parent = NULL;
2597 struct nfs_access_entry *entry;
2600 spin_lock(&inode->i_lock);
2601 while (*p != NULL) {
2603 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2604 cmp = cred_fscmp(set->cred, entry->cred);
2607 p = &parent->rb_left;
2609 p = &parent->rb_right;
2613 rb_link_node(&set->rb_node, parent, p);
2614 rb_insert_color(&set->rb_node, root_node);
2615 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2616 spin_unlock(&inode->i_lock);
2619 rb_replace_node(parent, &set->rb_node, root_node);
2620 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2621 list_del(&entry->lru);
2622 spin_unlock(&inode->i_lock);
2623 nfs_access_free_entry(entry);
2626 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2628 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2631 RB_CLEAR_NODE(&cache->rb_node);
2632 cache->cred = get_cred(set->cred);
2633 cache->mask = set->mask;
2635 /* The above field assignments must be visible
2636 * before this item appears on the lru. We cannot easily
2637 * use rcu_assign_pointer, so just force the memory barrier.
2640 nfs_access_add_rbtree(inode, cache);
2642 /* Update accounting */
2643 smp_mb__before_atomic();
2644 atomic_long_inc(&nfs_access_nr_entries);
2645 smp_mb__after_atomic();
2647 /* Add inode to global LRU list */
2648 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2649 spin_lock(&nfs_access_lru_lock);
2650 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2651 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2652 &nfs_access_lru_list);
2653 spin_unlock(&nfs_access_lru_lock);
2655 nfs_access_cache_enforce_limit();
2657 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2659 #define NFS_MAY_READ (NFS_ACCESS_READ)
2660 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2661 NFS_ACCESS_EXTEND | \
2663 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2665 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2666 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2667 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2669 nfs_access_calc_mask(u32 access_result, umode_t umode)
2673 if (access_result & NFS_MAY_READ)
2675 if (S_ISDIR(umode)) {
2676 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2678 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2680 } else if (S_ISREG(umode)) {
2681 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2683 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2685 } else if (access_result & NFS_MAY_WRITE)
2690 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2692 entry->mask = access_result;
2694 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2696 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2698 struct nfs_access_entry cache;
2699 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2700 int cache_mask = -1;
2703 trace_nfs_access_enter(inode);
2705 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2714 * Determine which access bits we want to ask for...
2716 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2717 if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2718 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2721 if (S_ISDIR(inode->i_mode))
2722 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2724 cache.mask |= NFS_ACCESS_EXECUTE;
2726 status = NFS_PROTO(inode)->access(inode, &cache);
2728 if (status == -ESTALE) {
2729 if (!S_ISDIR(inode->i_mode))
2730 nfs_set_inode_stale(inode);
2732 nfs_zap_caches(inode);
2736 nfs_access_add_cache(inode, &cache);
2738 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2739 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2742 trace_nfs_access_exit(inode, mask, cache_mask, status);
2746 static int nfs_open_permission_mask(int openflags)
2750 if (openflags & __FMODE_EXEC) {
2751 /* ONLY check exec rights */
2754 if ((openflags & O_ACCMODE) != O_WRONLY)
2756 if ((openflags & O_ACCMODE) != O_RDONLY)
2763 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2765 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2767 EXPORT_SYMBOL_GPL(nfs_may_open);
2769 static int nfs_execute_ok(struct inode *inode, int mask)
2771 struct nfs_server *server = NFS_SERVER(inode);
2774 if (S_ISDIR(inode->i_mode))
2776 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2777 if (mask & MAY_NOT_BLOCK)
2779 ret = __nfs_revalidate_inode(server, inode);
2781 if (ret == 0 && !execute_ok(inode))
2786 int nfs_permission(struct inode *inode, int mask)
2788 const struct cred *cred = current_cred();
2791 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2793 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2795 /* Is this sys_access() ? */
2796 if (mask & (MAY_ACCESS | MAY_CHDIR))
2799 switch (inode->i_mode & S_IFMT) {
2803 if ((mask & MAY_OPEN) &&
2804 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2809 * Optimize away all write operations, since the server
2810 * will check permissions when we perform the op.
2812 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2817 if (!NFS_PROTO(inode)->access)
2820 res = nfs_do_access(inode, cred, mask);
2822 if (!res && (mask & MAY_EXEC))
2823 res = nfs_execute_ok(inode, mask);
2825 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2826 inode->i_sb->s_id, inode->i_ino, mask, res);
2829 if (mask & MAY_NOT_BLOCK)
2832 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2834 res = generic_permission(inode, mask);
2837 EXPORT_SYMBOL_GPL(nfs_permission);
2841 * version-control: t
2842 * kept-new-versions: 5