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
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 {
144 struct nfs_cache_array_entry array[0];
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool);
151 struct dir_context *ctx;
152 unsigned long page_index;
155 loff_t current_index;
156 decode_dirent_t decode;
158 unsigned long dir_verifier;
159 unsigned long timestamp;
160 unsigned long gencount;
161 unsigned int cache_entry_index;
164 } nfs_readdir_descriptor_t;
167 void nfs_readdir_init_array(struct page *page)
169 struct nfs_cache_array *array;
171 array = kmap_atomic(page);
172 memset(array, 0, sizeof(struct nfs_cache_array));
173 array->eof_index = -1;
174 kunmap_atomic(array);
178 * we are freeing strings created by nfs_add_to_readdir_array()
181 void nfs_readdir_clear_array(struct page *page)
183 struct nfs_cache_array *array;
186 array = kmap_atomic(page);
187 for (i = 0; i < array->size; i++)
188 kfree(array->array[i].string.name);
190 kunmap_atomic(array);
194 * the caller is responsible for freeing qstr.name
195 * when called by nfs_readdir_add_to_array, the strings will be freed in
196 * nfs_clear_readdir_array()
199 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
202 string->name = kmemdup_nul(name, len, GFP_KERNEL);
203 if (string->name == NULL)
206 * Avoid a kmemleak false positive. The pointer to the name is stored
207 * in a page cache page which kmemleak does not scan.
209 kmemleak_not_leak(string->name);
210 string->hash = full_name_hash(NULL, name, len);
215 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
217 struct nfs_cache_array *array = kmap(page);
218 struct nfs_cache_array_entry *cache_entry;
221 cache_entry = &array->array[array->size];
223 /* Check that this entry lies within the page bounds */
225 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
228 cache_entry->cookie = entry->prev_cookie;
229 cache_entry->ino = entry->ino;
230 cache_entry->d_type = entry->d_type;
231 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
234 array->last_cookie = entry->cookie;
237 array->eof_index = array->size;
244 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
246 loff_t diff = desc->ctx->pos - desc->current_index;
251 if (diff >= array->size) {
252 if (array->eof_index >= 0)
257 index = (unsigned int)diff;
258 *desc->dir_cookie = array->array[index].cookie;
259 desc->cache_entry_index = index;
267 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
269 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
272 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
276 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
280 int status = -EAGAIN;
282 for (i = 0; i < array->size; i++) {
283 if (array->array[i].cookie == *desc->dir_cookie) {
284 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
285 struct nfs_open_dir_context *ctx = desc->file->private_data;
287 new_pos = desc->current_index + i;
288 if (ctx->attr_gencount != nfsi->attr_gencount ||
289 !nfs_readdir_inode_mapping_valid(nfsi)) {
291 ctx->attr_gencount = nfsi->attr_gencount;
292 } else if (new_pos < desc->ctx->pos) {
294 && ctx->dup_cookie == *desc->dir_cookie) {
295 if (printk_ratelimit()) {
296 pr_notice("NFS: directory %pD2 contains a readdir loop."
297 "Please contact your server vendor. "
298 "The file: %.*s has duplicate cookie %llu\n",
299 desc->file, array->array[i].string.len,
300 array->array[i].string.name, *desc->dir_cookie);
305 ctx->dup_cookie = *desc->dir_cookie;
308 desc->ctx->pos = new_pos;
309 desc->cache_entry_index = i;
313 if (array->eof_index >= 0) {
314 status = -EBADCOOKIE;
315 if (*desc->dir_cookie == array->last_cookie)
323 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
325 struct nfs_cache_array *array;
328 array = kmap(desc->page);
330 if (*desc->dir_cookie == 0)
331 status = nfs_readdir_search_for_pos(array, desc);
333 status = nfs_readdir_search_for_cookie(array, desc);
335 if (status == -EAGAIN) {
336 desc->last_cookie = array->last_cookie;
337 desc->current_index += array->size;
344 /* Fill a page with xdr information before transferring to the cache page */
346 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
347 struct nfs_entry *entry, struct file *file, struct inode *inode)
349 struct nfs_open_dir_context *ctx = file->private_data;
350 const struct cred *cred = ctx->cred;
351 unsigned long timestamp, gencount;
356 gencount = nfs_inc_attr_generation_counter();
357 desc->dir_verifier = nfs_save_change_attribute(inode);
358 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
359 NFS_SERVER(inode)->dtsize, desc->plus);
361 /* We requested READDIRPLUS, but the server doesn't grok it */
362 if (error == -ENOTSUPP && desc->plus) {
363 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
364 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
370 desc->timestamp = timestamp;
371 desc->gencount = gencount;
376 static int xdr_decode(nfs_readdir_descriptor_t *desc,
377 struct nfs_entry *entry, struct xdr_stream *xdr)
381 error = desc->decode(xdr, entry, desc->plus);
384 entry->fattr->time_start = desc->timestamp;
385 entry->fattr->gencount = desc->gencount;
389 /* Match file and dirent using either filehandle or fileid
390 * Note: caller is responsible for checking the fsid
393 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
396 struct nfs_inode *nfsi;
398 if (d_really_is_negative(dentry))
401 inode = d_inode(dentry);
402 if (is_bad_inode(inode) || NFS_STALE(inode))
406 if (entry->fattr->fileid != nfsi->fileid)
408 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
414 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
416 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
418 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
426 * This function is called by the lookup and getattr code to request the
427 * use of readdirplus to accelerate any future lookups in the same
430 void nfs_advise_use_readdirplus(struct inode *dir)
432 struct nfs_inode *nfsi = NFS_I(dir);
434 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
435 !list_empty(&nfsi->open_files))
436 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
440 * This function is mainly for use by nfs_getattr().
442 * If this is an 'ls -l', we want to force use of readdirplus.
443 * Do this by checking if there is an active file descriptor
444 * and calling nfs_advise_use_readdirplus, then forcing a
447 void nfs_force_use_readdirplus(struct inode *dir)
449 struct nfs_inode *nfsi = NFS_I(dir);
451 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
452 !list_empty(&nfsi->open_files)) {
453 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
454 invalidate_mapping_pages(dir->i_mapping,
455 nfsi->page_index + 1, -1);
460 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
461 unsigned long dir_verifier)
463 struct qstr filename = QSTR_INIT(entry->name, entry->len);
464 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
465 struct dentry *dentry;
466 struct dentry *alias;
470 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
472 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
474 if (filename.len == 0)
476 /* Validate that the name doesn't contain any illegal '\0' */
477 if (strnlen(filename.name, filename.len) != filename.len)
480 if (strnchr(filename.name, filename.len, '/'))
482 if (filename.name[0] == '.') {
483 if (filename.len == 1)
485 if (filename.len == 2 && filename.name[1] == '.')
488 filename.hash = full_name_hash(parent, filename.name, filename.len);
490 dentry = d_lookup(parent, &filename);
493 dentry = d_alloc_parallel(parent, &filename, &wq);
497 if (!d_in_lookup(dentry)) {
498 /* Is there a mountpoint here? If so, just exit */
499 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
500 &entry->fattr->fsid))
502 if (nfs_same_file(dentry, entry)) {
503 if (!entry->fh->size)
505 nfs_set_verifier(dentry, dir_verifier);
506 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
508 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
511 d_invalidate(dentry);
517 if (!entry->fh->size) {
518 d_lookup_done(dentry);
522 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
523 alias = d_splice_alias(inode, dentry);
524 d_lookup_done(dentry);
531 nfs_set_verifier(dentry, dir_verifier);
536 /* Perform conversion from xdr to cache array */
538 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
539 struct page **xdr_pages, struct page *page, unsigned int buflen)
541 struct xdr_stream stream;
543 struct page *scratch;
544 struct nfs_cache_array *array;
545 unsigned int count = 0;
548 scratch = alloc_page(GFP_KERNEL);
555 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
556 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
559 status = xdr_decode(desc, entry, &stream);
561 if (status == -EAGAIN)
569 nfs_prime_dcache(file_dentry(desc->file), entry,
572 status = nfs_readdir_add_to_array(entry, page);
575 } while (!entry->eof);
578 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
580 array->eof_index = array->size;
590 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
593 for (i = 0; i < npages; i++)
598 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
599 * to nfs_readdir_free_pages()
602 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
606 for (i = 0; i < npages; i++) {
607 struct page *page = alloc_page(GFP_KERNEL);
615 nfs_readdir_free_pages(pages, i);
620 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
622 struct page *pages[NFS_MAX_READDIR_PAGES];
623 struct nfs_entry entry;
624 struct file *file = desc->file;
625 struct nfs_cache_array *array;
626 int status = -ENOMEM;
627 unsigned int array_size = ARRAY_SIZE(pages);
629 nfs_readdir_init_array(page);
631 entry.prev_cookie = 0;
632 entry.cookie = desc->last_cookie;
634 entry.fh = nfs_alloc_fhandle();
635 entry.fattr = nfs_alloc_fattr();
636 entry.server = NFS_SERVER(inode);
637 if (entry.fh == NULL || entry.fattr == NULL)
640 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
641 if (IS_ERR(entry.label)) {
642 status = PTR_ERR(entry.label);
648 status = nfs_readdir_alloc_pages(pages, array_size);
650 goto out_release_array;
653 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
658 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
660 if (status == -ENOSPC)
664 } while (array->eof_index < 0);
666 nfs_readdir_free_pages(pages, array_size);
669 nfs4_label_free(entry.label);
671 nfs_free_fattr(entry.fattr);
672 nfs_free_fhandle(entry.fh);
677 * Now we cache directories properly, by converting xdr information
678 * to an array that can be used for lookups later. This results in
679 * fewer cache pages, since we can store more information on each page.
680 * We only need to convert from xdr once so future lookups are much simpler
683 int nfs_readdir_filler(void *data, struct page* page)
685 nfs_readdir_descriptor_t *desc = data;
686 struct inode *inode = file_inode(desc->file);
689 ret = nfs_readdir_xdr_to_array(desc, page, inode);
692 SetPageUptodate(page);
694 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
695 /* Should never happen */
696 nfs_zap_mapping(inode, inode->i_mapping);
701 nfs_readdir_clear_array(page);
707 void cache_page_release(nfs_readdir_descriptor_t *desc)
709 put_page(desc->page);
714 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
716 return read_cache_page(desc->file->f_mapping, desc->page_index,
717 nfs_readdir_filler, desc);
721 * Returns 0 if desc->dir_cookie was found on page desc->page_index
722 * and locks the page to prevent removal from the page cache.
725 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
727 struct inode *inode = file_inode(desc->file);
728 struct nfs_inode *nfsi = NFS_I(inode);
731 desc->page = get_cache_page(desc);
732 if (IS_ERR(desc->page))
733 return PTR_ERR(desc->page);
734 res = lock_page_killable(desc->page);
738 if (desc->page->mapping != NULL) {
739 res = nfs_readdir_search_array(desc);
741 nfsi->page_index = desc->page_index;
745 unlock_page(desc->page);
747 cache_page_release(desc);
751 /* Search for desc->dir_cookie from the beginning of the page cache */
753 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
757 if (desc->page_index == 0) {
758 desc->current_index = 0;
759 desc->last_cookie = 0;
762 res = find_and_lock_cache_page(desc);
763 } while (res == -EAGAIN);
768 * Once we've found the start of the dirent within a page: fill 'er up...
771 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
773 struct file *file = desc->file;
776 struct nfs_cache_array *array = NULL;
777 struct nfs_open_dir_context *ctx = file->private_data;
779 array = kmap(desc->page);
780 for (i = desc->cache_entry_index; i < array->size; i++) {
781 struct nfs_cache_array_entry *ent;
783 ent = &array->array[i];
784 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
785 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
790 if (i < (array->size-1))
791 *desc->dir_cookie = array->array[i+1].cookie;
793 *desc->dir_cookie = array->last_cookie;
797 if (array->eof_index >= 0)
801 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
802 (unsigned long long)*desc->dir_cookie, res);
807 * If we cannot find a cookie in our cache, we suspect that this is
808 * because it points to a deleted file, so we ask the server to return
809 * whatever it thinks is the next entry. We then feed this to filldir.
810 * If all goes well, we should then be able to find our way round the
811 * cache on the next call to readdir_search_pagecache();
813 * NOTE: we cannot add the anonymous page to the pagecache because
814 * the data it contains might not be page aligned. Besides,
815 * we should already have a complete representation of the
816 * directory in the page cache by the time we get here.
819 int uncached_readdir(nfs_readdir_descriptor_t *desc)
821 struct page *page = NULL;
823 struct inode *inode = file_inode(desc->file);
824 struct nfs_open_dir_context *ctx = desc->file->private_data;
826 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
827 (unsigned long long)*desc->dir_cookie);
829 page = alloc_page(GFP_HIGHUSER);
835 desc->page_index = 0;
836 desc->last_cookie = *desc->dir_cookie;
840 status = nfs_readdir_xdr_to_array(desc, page, inode);
844 status = nfs_do_filldir(desc);
847 nfs_readdir_clear_array(desc->page);
848 cache_page_release(desc);
850 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
855 /* The file offset position represents the dirent entry number. A
856 last cookie cache takes care of the common case of reading the
859 static int nfs_readdir(struct file *file, struct dir_context *ctx)
861 struct dentry *dentry = file_dentry(file);
862 struct inode *inode = d_inode(dentry);
863 nfs_readdir_descriptor_t my_desc,
865 struct nfs_open_dir_context *dir_ctx = file->private_data;
868 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
869 file, (long long)ctx->pos);
870 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
873 * ctx->pos points to the dirent entry number.
874 * *desc->dir_cookie has the cookie for the next entry. We have
875 * to either find the entry with the appropriate number or
876 * revalidate the cookie.
878 memset(desc, 0, sizeof(*desc));
882 desc->dir_cookie = &dir_ctx->dir_cookie;
883 desc->decode = NFS_PROTO(inode)->decode_dirent;
884 desc->plus = nfs_use_readdirplus(inode, ctx);
886 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
887 res = nfs_revalidate_mapping(inode, file->f_mapping);
892 res = readdir_search_pagecache(desc);
894 if (res == -EBADCOOKIE) {
896 /* This means either end of directory */
897 if (*desc->dir_cookie && !desc->eof) {
898 /* Or that the server has 'lost' a cookie */
899 res = uncached_readdir(desc);
905 if (res == -ETOOSMALL && desc->plus) {
906 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
907 nfs_zap_caches(inode);
908 desc->page_index = 0;
916 res = nfs_do_filldir(desc);
917 unlock_page(desc->page);
918 cache_page_release(desc);
921 } while (!desc->eof);
925 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
929 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
931 struct inode *inode = file_inode(filp);
932 struct nfs_open_dir_context *dir_ctx = filp->private_data;
934 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
935 filp, offset, whence);
949 offset += filp->f_pos;
955 if (offset != filp->f_pos) {
956 filp->f_pos = offset;
957 dir_ctx->dir_cookie = 0;
965 * All directory operations under NFS are synchronous, so fsync()
966 * is a dummy operation.
968 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
971 struct inode *inode = file_inode(filp);
973 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
976 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
982 * nfs_force_lookup_revalidate - Mark the directory as having changed
983 * @dir: pointer to directory inode
985 * This forces the revalidation code in nfs_lookup_revalidate() to do a
986 * full lookup on all child dentries of 'dir' whenever a change occurs
987 * on the server that might have invalidated our dcache.
989 * Note that we reserve bit '0' as a tag to let us know when a dentry
990 * was revalidated while holding a delegation on its inode.
992 * The caller should be holding dir->i_lock
994 void nfs_force_lookup_revalidate(struct inode *dir)
996 NFS_I(dir)->cache_change_attribute += 2;
998 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1001 * nfs_verify_change_attribute - Detects NFS remote directory changes
1002 * @dir: pointer to parent directory inode
1003 * @verf: previously saved change attribute
1005 * Return "false" if the verifiers doesn't match the change attribute.
1006 * This would usually indicate that the directory contents have changed on
1007 * the server, and that any dentries need revalidating.
1009 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1011 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1014 static void nfs_set_verifier_delegated(unsigned long *verf)
1019 #if IS_ENABLED(CONFIG_NFS_V4)
1020 static void nfs_unset_verifier_delegated(unsigned long *verf)
1024 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1026 static bool nfs_test_verifier_delegated(unsigned long verf)
1031 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1033 return nfs_test_verifier_delegated(dentry->d_time);
1036 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1038 struct inode *inode = d_inode(dentry);
1040 if (!nfs_verifier_is_delegated(dentry) &&
1041 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1043 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1044 nfs_set_verifier_delegated(&verf);
1046 dentry->d_time = verf;
1050 * nfs_set_verifier - save a parent directory verifier in the dentry
1051 * @dentry: pointer to dentry
1052 * @verf: verifier to save
1054 * Saves the parent directory verifier in @dentry. If the inode has
1055 * a delegation, we also tag the dentry as having been revalidated
1056 * while holding a delegation so that we know we don't have to
1057 * look it up again after a directory change.
1059 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1062 spin_lock(&dentry->d_lock);
1063 nfs_set_verifier_locked(dentry, verf);
1064 spin_unlock(&dentry->d_lock);
1066 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1068 #if IS_ENABLED(CONFIG_NFS_V4)
1070 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1071 * @inode: pointer to inode
1073 * Iterates through the dentries in the inode alias list and clears
1074 * the tag used to indicate that the dentry has been revalidated
1075 * while holding a delegation.
1076 * This function is intended for use when the delegation is being
1077 * returned or revoked.
1079 void nfs_clear_verifier_delegated(struct inode *inode)
1081 struct dentry *alias;
1085 spin_lock(&inode->i_lock);
1086 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1087 spin_lock(&alias->d_lock);
1088 nfs_unset_verifier_delegated(&alias->d_time);
1089 spin_unlock(&alias->d_lock);
1091 spin_unlock(&inode->i_lock);
1093 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1094 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1097 * A check for whether or not the parent directory has changed.
1098 * In the case it has, we assume that the dentries are untrustworthy
1099 * and may need to be looked up again.
1100 * If rcu_walk prevents us from performing a full check, return 0.
1102 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1105 if (IS_ROOT(dentry))
1107 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1109 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1111 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1112 if (nfs_mapping_need_revalidate_inode(dir)) {
1115 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1118 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1124 * Use intent information to check whether or not we're going to do
1125 * an O_EXCL create using this path component.
1127 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1129 if (NFS_PROTO(dir)->version == 2)
1131 return flags & LOOKUP_EXCL;
1135 * Inode and filehandle revalidation for lookups.
1137 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1138 * or if the intent information indicates that we're about to open this
1139 * particular file and the "nocto" mount flag is not set.
1143 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1145 struct nfs_server *server = NFS_SERVER(inode);
1148 if (IS_AUTOMOUNT(inode))
1151 if (flags & LOOKUP_OPEN) {
1152 switch (inode->i_mode & S_IFMT) {
1154 /* A NFSv4 OPEN will revalidate later */
1155 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1159 if (server->flags & NFS_MOUNT_NOCTO)
1161 /* NFS close-to-open cache consistency validation */
1166 /* VFS wants an on-the-wire revalidation */
1167 if (flags & LOOKUP_REVAL)
1170 return (inode->i_nlink == 0) ? -ESTALE : 0;
1172 if (flags & LOOKUP_RCU)
1174 ret = __nfs_revalidate_inode(server, inode);
1181 * We judge how long we want to trust negative
1182 * dentries by looking at the parent inode mtime.
1184 * If parent mtime has changed, we revalidate, else we wait for a
1185 * period corresponding to the parent's attribute cache timeout value.
1187 * If LOOKUP_RCU prevents us from performing a full check, return 1
1188 * suggesting a reval is needed.
1190 * Note that when creating a new file, or looking up a rename target,
1191 * then it shouldn't be necessary to revalidate a negative dentry.
1194 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1197 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1199 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1201 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1205 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1206 struct inode *inode, int error)
1210 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1214 nfs_mark_for_revalidate(dir);
1215 if (inode && S_ISDIR(inode->i_mode)) {
1216 /* Purge readdir caches. */
1217 nfs_zap_caches(inode);
1219 * We can't d_drop the root of a disconnected tree:
1220 * its d_hash is on the s_anon list and d_drop() would hide
1221 * it from shrink_dcache_for_unmount(), leading to busy
1222 * inodes on unmount and further oopses.
1224 if (IS_ROOT(dentry))
1227 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1231 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1232 __func__, dentry, error);
1237 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1241 if (nfs_neg_need_reval(dir, dentry, flags)) {
1242 if (flags & LOOKUP_RCU)
1246 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1250 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1251 struct inode *inode)
1253 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1254 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1258 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1259 struct inode *inode)
1261 struct nfs_fh *fhandle;
1262 struct nfs_fattr *fattr;
1263 struct nfs4_label *label;
1264 unsigned long dir_verifier;
1268 fhandle = nfs_alloc_fhandle();
1269 fattr = nfs_alloc_fattr();
1270 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1271 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1274 dir_verifier = nfs_save_change_attribute(dir);
1275 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1283 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1289 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1291 if (nfs_refresh_inode(inode, fattr) < 0)
1294 nfs_setsecurity(inode, fattr, label);
1295 nfs_set_verifier(dentry, dir_verifier);
1297 /* set a readdirplus hint that we had a cache miss */
1298 nfs_force_use_readdirplus(dir);
1301 nfs_free_fattr(fattr);
1302 nfs_free_fhandle(fhandle);
1303 nfs4_label_free(label);
1304 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1308 * This is called every time the dcache has a lookup hit,
1309 * and we should check whether we can really trust that
1312 * NOTE! The hit can be a negative hit too, don't assume
1315 * If the parent directory is seen to have changed, we throw out the
1316 * cached dentry and do a new lookup.
1319 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1322 struct inode *inode;
1325 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1326 inode = d_inode(dentry);
1329 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1331 if (is_bad_inode(inode)) {
1332 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1337 if (nfs_verifier_is_delegated(dentry))
1338 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1340 /* Force a full look up iff the parent directory has changed */
1341 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1342 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1343 error = nfs_lookup_verify_inode(inode, flags);
1345 if (error == -ESTALE)
1346 nfs_zap_caches(dir);
1349 nfs_advise_use_readdirplus(dir);
1353 if (flags & LOOKUP_RCU)
1356 if (NFS_STALE(inode))
1359 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1360 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1361 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1364 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1366 if (flags & LOOKUP_RCU)
1368 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1372 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1373 int (*reval)(struct inode *, struct dentry *, unsigned int))
1375 struct dentry *parent;
1379 if (flags & LOOKUP_RCU) {
1380 parent = READ_ONCE(dentry->d_parent);
1381 dir = d_inode_rcu(parent);
1384 ret = reval(dir, dentry, flags);
1385 if (parent != READ_ONCE(dentry->d_parent))
1388 parent = dget_parent(dentry);
1389 ret = reval(d_inode(parent), dentry, flags);
1395 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1397 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1401 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1402 * when we don't really care about the dentry name. This is called when a
1403 * pathwalk ends on a dentry that was not found via a normal lookup in the
1404 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1406 * In this situation, we just want to verify that the inode itself is OK
1407 * since the dentry might have changed on the server.
1409 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1411 struct inode *inode = d_inode(dentry);
1415 * I believe we can only get a negative dentry here in the case of a
1416 * procfs-style symlink. Just assume it's correct for now, but we may
1417 * eventually need to do something more here.
1420 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1425 if (is_bad_inode(inode)) {
1426 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1431 error = nfs_lookup_verify_inode(inode, flags);
1432 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1433 __func__, inode->i_ino, error ? "invalid" : "valid");
1438 * This is called from dput() when d_count is going to 0.
1440 static int nfs_dentry_delete(const struct dentry *dentry)
1442 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1443 dentry, dentry->d_flags);
1445 /* Unhash any dentry with a stale inode */
1446 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1449 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1450 /* Unhash it, so that ->d_iput() would be called */
1453 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1454 /* Unhash it, so that ancestors of killed async unlink
1455 * files will be cleaned up during umount */
1462 /* Ensure that we revalidate inode->i_nlink */
1463 static void nfs_drop_nlink(struct inode *inode)
1465 spin_lock(&inode->i_lock);
1466 /* drop the inode if we're reasonably sure this is the last link */
1467 if (inode->i_nlink > 0)
1469 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1470 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1471 | NFS_INO_INVALID_CTIME
1472 | NFS_INO_INVALID_OTHER
1473 | NFS_INO_REVAL_FORCED;
1474 spin_unlock(&inode->i_lock);
1478 * Called when the dentry loses inode.
1479 * We use it to clean up silly-renamed files.
1481 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1483 if (S_ISDIR(inode->i_mode))
1484 /* drop any readdir cache as it could easily be old */
1485 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1487 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1488 nfs_complete_unlink(dentry, inode);
1489 nfs_drop_nlink(inode);
1494 static void nfs_d_release(struct dentry *dentry)
1496 /* free cached devname value, if it survived that far */
1497 if (unlikely(dentry->d_fsdata)) {
1498 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1501 kfree(dentry->d_fsdata);
1505 const struct dentry_operations nfs_dentry_operations = {
1506 .d_revalidate = nfs_lookup_revalidate,
1507 .d_weak_revalidate = nfs_weak_revalidate,
1508 .d_delete = nfs_dentry_delete,
1509 .d_iput = nfs_dentry_iput,
1510 .d_automount = nfs_d_automount,
1511 .d_release = nfs_d_release,
1513 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1515 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1518 struct inode *inode = NULL;
1519 struct nfs_fh *fhandle = NULL;
1520 struct nfs_fattr *fattr = NULL;
1521 struct nfs4_label *label = NULL;
1522 unsigned long dir_verifier;
1525 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1526 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1528 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1529 return ERR_PTR(-ENAMETOOLONG);
1532 * If we're doing an exclusive create, optimize away the lookup
1533 * but don't hash the dentry.
1535 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1538 res = ERR_PTR(-ENOMEM);
1539 fhandle = nfs_alloc_fhandle();
1540 fattr = nfs_alloc_fattr();
1541 if (fhandle == NULL || fattr == NULL)
1544 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1548 dir_verifier = nfs_save_change_attribute(dir);
1549 trace_nfs_lookup_enter(dir, dentry, flags);
1550 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1551 if (error == -ENOENT)
1554 res = ERR_PTR(error);
1557 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1558 res = ERR_CAST(inode);
1562 /* Notify readdir to use READDIRPLUS */
1563 nfs_force_use_readdirplus(dir);
1566 res = d_splice_alias(inode, dentry);
1572 nfs_set_verifier(dentry, dir_verifier);
1574 trace_nfs_lookup_exit(dir, dentry, flags, error);
1575 nfs4_label_free(label);
1577 nfs_free_fattr(fattr);
1578 nfs_free_fhandle(fhandle);
1581 EXPORT_SYMBOL_GPL(nfs_lookup);
1583 #if IS_ENABLED(CONFIG_NFS_V4)
1584 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1586 const struct dentry_operations nfs4_dentry_operations = {
1587 .d_revalidate = nfs4_lookup_revalidate,
1588 .d_weak_revalidate = nfs_weak_revalidate,
1589 .d_delete = nfs_dentry_delete,
1590 .d_iput = nfs_dentry_iput,
1591 .d_automount = nfs_d_automount,
1592 .d_release = nfs_d_release,
1594 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1596 static fmode_t flags_to_mode(int flags)
1598 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1599 if ((flags & O_ACCMODE) != O_WRONLY)
1601 if ((flags & O_ACCMODE) != O_RDONLY)
1606 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1608 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1611 static int do_open(struct inode *inode, struct file *filp)
1613 nfs_fscache_open_file(inode, filp);
1617 static int nfs_finish_open(struct nfs_open_context *ctx,
1618 struct dentry *dentry,
1619 struct file *file, unsigned open_flags)
1623 err = finish_open(file, dentry, do_open);
1626 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1627 nfs_file_set_open_context(file, ctx);
1634 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1635 struct file *file, unsigned open_flags,
1638 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1639 struct nfs_open_context *ctx;
1641 struct iattr attr = { .ia_valid = ATTR_OPEN };
1642 struct inode *inode;
1643 unsigned int lookup_flags = 0;
1644 bool switched = false;
1648 /* Expect a negative dentry */
1649 BUG_ON(d_inode(dentry));
1651 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1652 dir->i_sb->s_id, dir->i_ino, dentry);
1654 err = nfs_check_flags(open_flags);
1658 /* NFS only supports OPEN on regular files */
1659 if ((open_flags & O_DIRECTORY)) {
1660 if (!d_in_lookup(dentry)) {
1662 * Hashed negative dentry with O_DIRECTORY: dentry was
1663 * revalidated and is fine, no need to perform lookup
1668 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1672 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1673 return -ENAMETOOLONG;
1675 if (open_flags & O_CREAT) {
1676 struct nfs_server *server = NFS_SERVER(dir);
1678 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1679 mode &= ~current_umask();
1681 attr.ia_valid |= ATTR_MODE;
1682 attr.ia_mode = mode;
1684 if (open_flags & O_TRUNC) {
1685 attr.ia_valid |= ATTR_SIZE;
1689 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1692 dentry = d_alloc_parallel(dentry->d_parent,
1693 &dentry->d_name, &wq);
1695 return PTR_ERR(dentry);
1696 if (unlikely(!d_in_lookup(dentry)))
1697 return finish_no_open(file, dentry);
1700 ctx = create_nfs_open_context(dentry, open_flags, file);
1705 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1706 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1708 file->f_mode |= FMODE_CREATED;
1709 if (IS_ERR(inode)) {
1710 err = PTR_ERR(inode);
1711 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1712 put_nfs_open_context(ctx);
1716 d_splice_alias(NULL, dentry);
1717 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1723 if (!(open_flags & O_NOFOLLOW))
1733 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1734 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1735 put_nfs_open_context(ctx);
1737 if (unlikely(switched)) {
1738 d_lookup_done(dentry);
1744 res = nfs_lookup(dir, dentry, lookup_flags);
1746 d_lookup_done(dentry);
1753 return PTR_ERR(res);
1754 return finish_no_open(file, res);
1756 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1759 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1762 struct inode *inode;
1764 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1766 if (d_mountpoint(dentry))
1769 inode = d_inode(dentry);
1771 /* We can't create new files in nfs_open_revalidate(), so we
1772 * optimize away revalidation of negative dentries.
1777 if (nfs_verifier_is_delegated(dentry))
1778 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1780 /* NFS only supports OPEN on regular files */
1781 if (!S_ISREG(inode->i_mode))
1784 /* We cannot do exclusive creation on a positive dentry */
1785 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1788 /* Check if the directory changed */
1789 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1792 /* Let f_op->open() actually open (and revalidate) the file */
1795 if (flags & LOOKUP_RCU)
1797 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1800 return nfs_do_lookup_revalidate(dir, dentry, flags);
1803 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1805 return __nfs_lookup_revalidate(dentry, flags,
1806 nfs4_do_lookup_revalidate);
1809 #endif /* CONFIG_NFSV4 */
1812 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1813 struct nfs_fattr *fattr,
1814 struct nfs4_label *label)
1816 struct dentry *parent = dget_parent(dentry);
1817 struct inode *dir = d_inode(parent);
1818 struct inode *inode;
1824 if (fhandle->size == 0) {
1825 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
1829 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1830 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1831 struct nfs_server *server = NFS_SB(dentry->d_sb);
1832 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1837 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1838 d = d_splice_alias(inode, dentry);
1843 nfs_mark_for_revalidate(dir);
1847 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1850 * Code common to create, mkdir, and mknod.
1852 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1853 struct nfs_fattr *fattr,
1854 struct nfs4_label *label)
1858 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1862 /* Callers don't care */
1866 EXPORT_SYMBOL_GPL(nfs_instantiate);
1869 * Following a failed create operation, we drop the dentry rather
1870 * than retain a negative dentry. This avoids a problem in the event
1871 * that the operation succeeded on the server, but an error in the
1872 * reply path made it appear to have failed.
1874 int nfs_create(struct inode *dir, struct dentry *dentry,
1875 umode_t mode, bool excl)
1878 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1881 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1882 dir->i_sb->s_id, dir->i_ino, dentry);
1884 attr.ia_mode = mode;
1885 attr.ia_valid = ATTR_MODE;
1887 trace_nfs_create_enter(dir, dentry, open_flags);
1888 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1889 trace_nfs_create_exit(dir, dentry, open_flags, error);
1897 EXPORT_SYMBOL_GPL(nfs_create);
1900 * See comments for nfs_proc_create regarding failed operations.
1903 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1908 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1909 dir->i_sb->s_id, dir->i_ino, dentry);
1911 attr.ia_mode = mode;
1912 attr.ia_valid = ATTR_MODE;
1914 trace_nfs_mknod_enter(dir, dentry);
1915 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1916 trace_nfs_mknod_exit(dir, dentry, status);
1924 EXPORT_SYMBOL_GPL(nfs_mknod);
1927 * See comments for nfs_proc_create regarding failed operations.
1929 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1934 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1935 dir->i_sb->s_id, dir->i_ino, dentry);
1937 attr.ia_valid = ATTR_MODE;
1938 attr.ia_mode = mode | S_IFDIR;
1940 trace_nfs_mkdir_enter(dir, dentry);
1941 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1942 trace_nfs_mkdir_exit(dir, dentry, error);
1950 EXPORT_SYMBOL_GPL(nfs_mkdir);
1952 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1954 if (simple_positive(dentry))
1958 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1962 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1963 dir->i_sb->s_id, dir->i_ino, dentry);
1965 trace_nfs_rmdir_enter(dir, dentry);
1966 if (d_really_is_positive(dentry)) {
1967 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1968 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1969 /* Ensure the VFS deletes this inode */
1972 clear_nlink(d_inode(dentry));
1975 nfs_dentry_handle_enoent(dentry);
1977 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1979 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1980 trace_nfs_rmdir_exit(dir, dentry, error);
1984 EXPORT_SYMBOL_GPL(nfs_rmdir);
1987 * Remove a file after making sure there are no pending writes,
1988 * and after checking that the file has only one user.
1990 * We invalidate the attribute cache and free the inode prior to the operation
1991 * to avoid possible races if the server reuses the inode.
1993 static int nfs_safe_remove(struct dentry *dentry)
1995 struct inode *dir = d_inode(dentry->d_parent);
1996 struct inode *inode = d_inode(dentry);
1999 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2001 /* If the dentry was sillyrenamed, we simply call d_delete() */
2002 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2007 trace_nfs_remove_enter(dir, dentry);
2008 if (inode != NULL) {
2009 error = NFS_PROTO(dir)->remove(dir, dentry);
2011 nfs_drop_nlink(inode);
2013 error = NFS_PROTO(dir)->remove(dir, dentry);
2014 if (error == -ENOENT)
2015 nfs_dentry_handle_enoent(dentry);
2016 trace_nfs_remove_exit(dir, dentry, error);
2021 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2022 * belongs to an active ".nfs..." file and we return -EBUSY.
2024 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2026 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2029 int need_rehash = 0;
2031 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2032 dir->i_ino, dentry);
2034 trace_nfs_unlink_enter(dir, dentry);
2035 spin_lock(&dentry->d_lock);
2036 if (d_count(dentry) > 1) {
2037 spin_unlock(&dentry->d_lock);
2038 /* Start asynchronous writeout of the inode */
2039 write_inode_now(d_inode(dentry), 0);
2040 error = nfs_sillyrename(dir, dentry);
2043 if (!d_unhashed(dentry)) {
2047 spin_unlock(&dentry->d_lock);
2048 error = nfs_safe_remove(dentry);
2049 if (!error || error == -ENOENT) {
2050 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2051 } else if (need_rehash)
2054 trace_nfs_unlink_exit(dir, dentry, error);
2057 EXPORT_SYMBOL_GPL(nfs_unlink);
2060 * To create a symbolic link, most file systems instantiate a new inode,
2061 * add a page to it containing the path, then write it out to the disk
2062 * using prepare_write/commit_write.
2064 * Unfortunately the NFS client can't create the in-core inode first
2065 * because it needs a file handle to create an in-core inode (see
2066 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2067 * symlink request has completed on the server.
2069 * So instead we allocate a raw page, copy the symname into it, then do
2070 * the SYMLINK request with the page as the buffer. If it succeeds, we
2071 * now have a new file handle and can instantiate an in-core NFS inode
2072 * and move the raw page into its mapping.
2074 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2079 unsigned int pathlen = strlen(symname);
2082 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2083 dir->i_ino, dentry, symname);
2085 if (pathlen > PAGE_SIZE)
2086 return -ENAMETOOLONG;
2088 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2089 attr.ia_valid = ATTR_MODE;
2091 page = alloc_page(GFP_USER);
2095 kaddr = page_address(page);
2096 memcpy(kaddr, symname, pathlen);
2097 if (pathlen < PAGE_SIZE)
2098 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2100 trace_nfs_symlink_enter(dir, dentry);
2101 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2102 trace_nfs_symlink_exit(dir, dentry, error);
2104 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2105 dir->i_sb->s_id, dir->i_ino,
2106 dentry, symname, error);
2113 * No big deal if we can't add this page to the page cache here.
2114 * READLINK will get the missing page from the server if needed.
2116 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2118 SetPageUptodate(page);
2121 * add_to_page_cache_lru() grabs an extra page refcount.
2122 * Drop it here to avoid leaking this page later.
2130 EXPORT_SYMBOL_GPL(nfs_symlink);
2133 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2135 struct inode *inode = d_inode(old_dentry);
2138 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2139 old_dentry, dentry);
2141 trace_nfs_link_enter(inode, dir, dentry);
2143 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2146 d_add(dentry, inode);
2148 trace_nfs_link_exit(inode, dir, dentry, error);
2151 EXPORT_SYMBOL_GPL(nfs_link);
2155 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2156 * different file handle for the same inode after a rename (e.g. when
2157 * moving to a different directory). A fail-safe method to do so would
2158 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2159 * rename the old file using the sillyrename stuff. This way, the original
2160 * file in old_dir will go away when the last process iput()s the inode.
2164 * It actually works quite well. One needs to have the possibility for
2165 * at least one ".nfs..." file in each directory the file ever gets
2166 * moved or linked to which happens automagically with the new
2167 * implementation that only depends on the dcache stuff instead of
2168 * using the inode layer
2170 * Unfortunately, things are a little more complicated than indicated
2171 * above. For a cross-directory move, we want to make sure we can get
2172 * rid of the old inode after the operation. This means there must be
2173 * no pending writes (if it's a file), and the use count must be 1.
2174 * If these conditions are met, we can drop the dentries before doing
2177 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2178 struct inode *new_dir, struct dentry *new_dentry,
2181 struct inode *old_inode = d_inode(old_dentry);
2182 struct inode *new_inode = d_inode(new_dentry);
2183 struct dentry *dentry = NULL, *rehash = NULL;
2184 struct rpc_task *task;
2190 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2191 old_dentry, new_dentry,
2192 d_count(new_dentry));
2194 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2196 * For non-directories, check whether the target is busy and if so,
2197 * make a copy of the dentry and then do a silly-rename. If the
2198 * silly-rename succeeds, the copied dentry is hashed and becomes
2201 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2203 * To prevent any new references to the target during the
2204 * rename, we unhash the dentry in advance.
2206 if (!d_unhashed(new_dentry)) {
2208 rehash = new_dentry;
2211 if (d_count(new_dentry) > 2) {
2214 /* copy the target dentry's name */
2215 dentry = d_alloc(new_dentry->d_parent,
2216 &new_dentry->d_name);
2220 /* silly-rename the existing target ... */
2221 err = nfs_sillyrename(new_dir, new_dentry);
2225 new_dentry = dentry;
2231 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2233 error = PTR_ERR(task);
2237 error = rpc_wait_for_completion_task(task);
2239 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2240 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2243 error = task->tk_status;
2245 /* Ensure the inode attributes are revalidated */
2247 spin_lock(&old_inode->i_lock);
2248 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2249 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2250 | NFS_INO_INVALID_CTIME
2251 | NFS_INO_REVAL_FORCED;
2252 spin_unlock(&old_inode->i_lock);
2257 trace_nfs_rename_exit(old_dir, old_dentry,
2258 new_dir, new_dentry, error);
2260 if (new_inode != NULL)
2261 nfs_drop_nlink(new_inode);
2263 * The d_move() should be here instead of in an async RPC completion
2264 * handler because we need the proper locks to move the dentry. If
2265 * we're interrupted by a signal, the async RPC completion handler
2266 * should mark the directories for revalidation.
2268 d_move(old_dentry, new_dentry);
2269 nfs_set_verifier(old_dentry,
2270 nfs_save_change_attribute(new_dir));
2271 } else if (error == -ENOENT)
2272 nfs_dentry_handle_enoent(old_dentry);
2274 /* new dentry created? */
2279 EXPORT_SYMBOL_GPL(nfs_rename);
2281 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2282 static LIST_HEAD(nfs_access_lru_list);
2283 static atomic_long_t nfs_access_nr_entries;
2285 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2286 module_param(nfs_access_max_cachesize, ulong, 0644);
2287 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2289 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2291 put_cred(entry->cred);
2292 kfree_rcu(entry, rcu_head);
2293 smp_mb__before_atomic();
2294 atomic_long_dec(&nfs_access_nr_entries);
2295 smp_mb__after_atomic();
2298 static void nfs_access_free_list(struct list_head *head)
2300 struct nfs_access_entry *cache;
2302 while (!list_empty(head)) {
2303 cache = list_entry(head->next, struct nfs_access_entry, lru);
2304 list_del(&cache->lru);
2305 nfs_access_free_entry(cache);
2309 static unsigned long
2310 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2313 struct nfs_inode *nfsi, *next;
2314 struct nfs_access_entry *cache;
2317 spin_lock(&nfs_access_lru_lock);
2318 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2319 struct inode *inode;
2321 if (nr_to_scan-- == 0)
2323 inode = &nfsi->vfs_inode;
2324 spin_lock(&inode->i_lock);
2325 if (list_empty(&nfsi->access_cache_entry_lru))
2326 goto remove_lru_entry;
2327 cache = list_entry(nfsi->access_cache_entry_lru.next,
2328 struct nfs_access_entry, lru);
2329 list_move(&cache->lru, &head);
2330 rb_erase(&cache->rb_node, &nfsi->access_cache);
2332 if (!list_empty(&nfsi->access_cache_entry_lru))
2333 list_move_tail(&nfsi->access_cache_inode_lru,
2334 &nfs_access_lru_list);
2337 list_del_init(&nfsi->access_cache_inode_lru);
2338 smp_mb__before_atomic();
2339 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2340 smp_mb__after_atomic();
2342 spin_unlock(&inode->i_lock);
2344 spin_unlock(&nfs_access_lru_lock);
2345 nfs_access_free_list(&head);
2350 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2352 int nr_to_scan = sc->nr_to_scan;
2353 gfp_t gfp_mask = sc->gfp_mask;
2355 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2357 return nfs_do_access_cache_scan(nr_to_scan);
2362 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2364 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2368 nfs_access_cache_enforce_limit(void)
2370 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2372 unsigned int nr_to_scan;
2374 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2377 diff = nr_entries - nfs_access_max_cachesize;
2378 if (diff < nr_to_scan)
2380 nfs_do_access_cache_scan(nr_to_scan);
2383 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2385 struct rb_root *root_node = &nfsi->access_cache;
2387 struct nfs_access_entry *entry;
2389 /* Unhook entries from the cache */
2390 while ((n = rb_first(root_node)) != NULL) {
2391 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2392 rb_erase(n, root_node);
2393 list_move(&entry->lru, head);
2395 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2398 void nfs_access_zap_cache(struct inode *inode)
2402 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2404 /* Remove from global LRU init */
2405 spin_lock(&nfs_access_lru_lock);
2406 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2407 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2409 spin_lock(&inode->i_lock);
2410 __nfs_access_zap_cache(NFS_I(inode), &head);
2411 spin_unlock(&inode->i_lock);
2412 spin_unlock(&nfs_access_lru_lock);
2413 nfs_access_free_list(&head);
2415 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2417 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2419 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2422 struct nfs_access_entry *entry =
2423 rb_entry(n, struct nfs_access_entry, rb_node);
2424 int cmp = cred_fscmp(cred, entry->cred);
2436 static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2438 struct nfs_inode *nfsi = NFS_I(inode);
2439 struct nfs_access_entry *cache;
2443 spin_lock(&inode->i_lock);
2445 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2447 cache = nfs_access_search_rbtree(inode, cred);
2451 /* Found an entry, is our attribute cache valid? */
2452 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2459 spin_unlock(&inode->i_lock);
2460 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2463 spin_lock(&inode->i_lock);
2466 res->cred = cache->cred;
2467 res->mask = cache->mask;
2468 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2471 spin_unlock(&inode->i_lock);
2474 spin_unlock(&inode->i_lock);
2475 nfs_access_zap_cache(inode);
2479 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2481 /* Only check the most recently returned cache entry,
2482 * but do it without locking.
2484 struct nfs_inode *nfsi = NFS_I(inode);
2485 struct nfs_access_entry *cache;
2487 struct list_head *lh;
2490 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2492 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2493 cache = list_entry(lh, struct nfs_access_entry, lru);
2494 if (lh == &nfsi->access_cache_entry_lru ||
2495 cred_fscmp(cred, cache->cred) != 0)
2499 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2501 res->cred = cache->cred;
2502 res->mask = cache->mask;
2509 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2511 struct nfs_inode *nfsi = NFS_I(inode);
2512 struct rb_root *root_node = &nfsi->access_cache;
2513 struct rb_node **p = &root_node->rb_node;
2514 struct rb_node *parent = NULL;
2515 struct nfs_access_entry *entry;
2518 spin_lock(&inode->i_lock);
2519 while (*p != NULL) {
2521 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2522 cmp = cred_fscmp(set->cred, entry->cred);
2525 p = &parent->rb_left;
2527 p = &parent->rb_right;
2531 rb_link_node(&set->rb_node, parent, p);
2532 rb_insert_color(&set->rb_node, root_node);
2533 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2534 spin_unlock(&inode->i_lock);
2537 rb_replace_node(parent, &set->rb_node, root_node);
2538 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2539 list_del(&entry->lru);
2540 spin_unlock(&inode->i_lock);
2541 nfs_access_free_entry(entry);
2544 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2546 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2549 RB_CLEAR_NODE(&cache->rb_node);
2550 cache->cred = get_cred(set->cred);
2551 cache->mask = set->mask;
2553 /* The above field assignments must be visible
2554 * before this item appears on the lru. We cannot easily
2555 * use rcu_assign_pointer, so just force the memory barrier.
2558 nfs_access_add_rbtree(inode, cache);
2560 /* Update accounting */
2561 smp_mb__before_atomic();
2562 atomic_long_inc(&nfs_access_nr_entries);
2563 smp_mb__after_atomic();
2565 /* Add inode to global LRU list */
2566 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2567 spin_lock(&nfs_access_lru_lock);
2568 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2569 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2570 &nfs_access_lru_list);
2571 spin_unlock(&nfs_access_lru_lock);
2573 nfs_access_cache_enforce_limit();
2575 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2577 #define NFS_MAY_READ (NFS_ACCESS_READ)
2578 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2579 NFS_ACCESS_EXTEND | \
2581 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2583 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2584 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2585 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2587 nfs_access_calc_mask(u32 access_result, umode_t umode)
2591 if (access_result & NFS_MAY_READ)
2593 if (S_ISDIR(umode)) {
2594 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2596 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2598 } else if (S_ISREG(umode)) {
2599 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2601 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2603 } else if (access_result & NFS_MAY_WRITE)
2608 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2610 entry->mask = access_result;
2612 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2614 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2616 struct nfs_access_entry cache;
2617 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2618 int cache_mask = -1;
2621 trace_nfs_access_enter(inode);
2623 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2625 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2634 * Determine which access bits we want to ask for...
2636 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2637 if (S_ISDIR(inode->i_mode))
2638 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2640 cache.mask |= NFS_ACCESS_EXECUTE;
2642 status = NFS_PROTO(inode)->access(inode, &cache);
2644 if (status == -ESTALE) {
2645 nfs_zap_caches(inode);
2646 if (!S_ISDIR(inode->i_mode))
2647 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2651 nfs_access_add_cache(inode, &cache);
2653 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2654 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2657 trace_nfs_access_exit(inode, mask, cache_mask, status);
2661 static int nfs_open_permission_mask(int openflags)
2665 if (openflags & __FMODE_EXEC) {
2666 /* ONLY check exec rights */
2669 if ((openflags & O_ACCMODE) != O_WRONLY)
2671 if ((openflags & O_ACCMODE) != O_RDONLY)
2678 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2680 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2682 EXPORT_SYMBOL_GPL(nfs_may_open);
2684 static int nfs_execute_ok(struct inode *inode, int mask)
2686 struct nfs_server *server = NFS_SERVER(inode);
2689 if (S_ISDIR(inode->i_mode))
2691 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2692 if (mask & MAY_NOT_BLOCK)
2694 ret = __nfs_revalidate_inode(server, inode);
2696 if (ret == 0 && !execute_ok(inode))
2701 int nfs_permission(struct inode *inode, int mask)
2703 const struct cred *cred = current_cred();
2706 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2708 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2710 /* Is this sys_access() ? */
2711 if (mask & (MAY_ACCESS | MAY_CHDIR))
2714 switch (inode->i_mode & S_IFMT) {
2718 if ((mask & MAY_OPEN) &&
2719 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2724 * Optimize away all write operations, since the server
2725 * will check permissions when we perform the op.
2727 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2732 if (!NFS_PROTO(inode)->access)
2735 /* Always try fast lookups first */
2737 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2739 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2740 /* Fast lookup failed, try the slow way */
2741 res = nfs_do_access(inode, cred, mask);
2744 if (!res && (mask & MAY_EXEC))
2745 res = nfs_execute_ok(inode, mask);
2747 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2748 inode->i_sb->s_id, inode->i_ino, mask, res);
2751 if (mask & MAY_NOT_BLOCK)
2754 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2756 res = generic_permission(inode, mask);
2759 EXPORT_SYMBOL_GPL(nfs_permission);
2763 * version-control: t
2764 * kept-new-versions: 5