4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
23 #include <linux/stat.h>
24 #include <linux/fcntl.h>
25 #include <linux/string.h>
26 #include <linux/kernel.h>
27 #include <linux/slab.h>
29 #include <linux/sunrpc/clnt.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/pagemap.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
36 #include <linux/swap.h>
37 #include <linux/sched.h>
38 #include <linux/kmemleak.h>
39 #include <linux/xattr.h>
41 #include "delegation.h"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
57 const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .iterate = nfs_readdir,
62 .release = nfs_closedir,
63 .fsync = nfs_fsync_dir,
66 const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
72 struct nfs_inode *nfsi = NFS_I(dir);
73 struct nfs_open_dir_context *ctx;
74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
77 ctx->attr_gencount = nfsi->attr_gencount;
80 ctx->cred = get_rpccred(cred);
81 spin_lock(&dir->i_lock);
82 list_add(&ctx->list, &nfsi->open_files);
83 spin_unlock(&dir->i_lock);
86 return ERR_PTR(-ENOMEM);
89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
91 spin_lock(&dir->i_lock);
93 spin_unlock(&dir->i_lock);
94 put_rpccred(ctx->cred);
102 nfs_opendir(struct inode *inode, struct file *filp)
105 struct nfs_open_dir_context *ctx;
106 struct rpc_cred *cred;
108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
110 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
112 cred = rpc_lookup_cred();
114 return PTR_ERR(cred);
115 ctx = alloc_nfs_open_dir_context(inode, cred);
120 filp->private_data = ctx;
121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 /* This is a mountpoint, so d_revalidate will never
123 * have been called, so we need to refresh the
124 * inode (for close-open consistency) ourselves.
126 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
134 nfs_closedir(struct inode *inode, struct file *filp)
136 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
140 struct nfs_cache_array_entry {
144 unsigned char d_type;
147 struct nfs_cache_array {
151 struct nfs_cache_array_entry array[0];
154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
158 struct dir_context *ctx;
159 unsigned long page_index;
162 loff_t current_index;
163 decode_dirent_t decode;
165 unsigned long timestamp;
166 unsigned long gencount;
167 unsigned int cache_entry_index;
170 } nfs_readdir_descriptor_t;
173 * The caller is responsible for calling nfs_readdir_release_array(page)
176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
180 return ERR_PTR(-EIO);
183 return ERR_PTR(-ENOMEM);
188 void nfs_readdir_release_array(struct page *page)
194 * we are freeing strings created by nfs_add_to_readdir_array()
197 void nfs_readdir_clear_array(struct page *page)
199 struct nfs_cache_array *array;
202 array = kmap_atomic(page);
203 for (i = 0; i < array->size; i++)
204 kfree(array->array[i].string.name);
205 kunmap_atomic(array);
209 * the caller is responsible for freeing qstr.name
210 * when called by nfs_readdir_add_to_array, the strings will be freed in
211 * nfs_clear_readdir_array()
214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
217 string->name = kmemdup(name, len, GFP_KERNEL);
218 if (string->name == NULL)
221 * Avoid a kmemleak false positive. The pointer to the name is stored
222 * in a page cache page which kmemleak does not scan.
224 kmemleak_not_leak(string->name);
225 string->hash = full_name_hash(name, len);
230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
232 struct nfs_cache_array *array = nfs_readdir_get_array(page);
233 struct nfs_cache_array_entry *cache_entry;
237 return PTR_ERR(array);
239 cache_entry = &array->array[array->size];
241 /* Check that this entry lies within the page bounds */
243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
246 cache_entry->cookie = entry->prev_cookie;
247 cache_entry->ino = entry->ino;
248 cache_entry->d_type = entry->d_type;
249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
252 array->last_cookie = entry->cookie;
255 array->eof_index = array->size;
257 nfs_readdir_release_array(page);
262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
264 loff_t diff = desc->ctx->pos - desc->current_index;
269 if (diff >= array->size) {
270 if (array->eof_index >= 0)
275 index = (unsigned int)diff;
276 *desc->dir_cookie = array->array[index].cookie;
277 desc->cache_entry_index = index;
285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
298 int status = -EAGAIN;
300 for (i = 0; i < array->size; i++) {
301 if (array->array[i].cookie == *desc->dir_cookie) {
302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303 struct nfs_open_dir_context *ctx = desc->file->private_data;
305 new_pos = desc->current_index + i;
306 if (ctx->attr_gencount != nfsi->attr_gencount ||
307 !nfs_readdir_inode_mapping_valid(nfsi)) {
309 ctx->attr_gencount = nfsi->attr_gencount;
310 } else if (new_pos < desc->ctx->pos) {
312 && ctx->dup_cookie == *desc->dir_cookie) {
313 if (printk_ratelimit()) {
314 pr_notice("NFS: directory %pD2 contains a readdir loop."
315 "Please contact your server vendor. "
316 "The file: %.*s has duplicate cookie %llu\n",
317 desc->file, array->array[i].string.len,
318 array->array[i].string.name, *desc->dir_cookie);
323 ctx->dup_cookie = *desc->dir_cookie;
326 desc->ctx->pos = new_pos;
327 desc->cache_entry_index = i;
331 if (array->eof_index >= 0) {
332 status = -EBADCOOKIE;
333 if (*desc->dir_cookie == array->last_cookie)
341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
343 struct nfs_cache_array *array;
346 array = nfs_readdir_get_array(desc->page);
348 status = PTR_ERR(array);
352 if (*desc->dir_cookie == 0)
353 status = nfs_readdir_search_for_pos(array, desc);
355 status = nfs_readdir_search_for_cookie(array, desc);
357 if (status == -EAGAIN) {
358 desc->last_cookie = array->last_cookie;
359 desc->current_index += array->size;
362 nfs_readdir_release_array(desc->page);
367 /* Fill a page with xdr information before transferring to the cache page */
369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
370 struct nfs_entry *entry, struct file *file, struct inode *inode)
372 struct nfs_open_dir_context *ctx = file->private_data;
373 struct rpc_cred *cred = ctx->cred;
374 unsigned long timestamp, gencount;
379 gencount = nfs_inc_attr_generation_counter();
380 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
381 NFS_SERVER(inode)->dtsize, desc->plus);
383 /* We requested READDIRPLUS, but the server doesn't grok it */
384 if (error == -ENOTSUPP && desc->plus) {
385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
392 desc->timestamp = timestamp;
393 desc->gencount = gencount;
398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399 struct nfs_entry *entry, struct xdr_stream *xdr)
403 error = desc->decode(xdr, entry, desc->plus);
406 entry->fattr->time_start = desc->timestamp;
407 entry->fattr->gencount = desc->gencount;
411 /* Match file and dirent using either filehandle or fileid
412 * Note: caller is responsible for checking the fsid
415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
417 struct nfs_inode *nfsi;
419 if (dentry->d_inode == NULL)
422 nfsi = NFS_I(dentry->d_inode);
423 if (entry->fattr->fileid == nfsi->fileid)
425 if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0)
432 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
434 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
436 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
444 * This function is called by the lookup code to request the use of
445 * readdirplus to accelerate any future lookups in the same
449 void nfs_advise_use_readdirplus(struct inode *dir)
451 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
455 * This function is mainly for use by nfs_getattr().
457 * If this is an 'ls -l', we want to force use of readdirplus.
458 * Do this by checking if there is an active file descriptor
459 * and calling nfs_advise_use_readdirplus, then forcing a
462 void nfs_force_use_readdirplus(struct inode *dir)
464 if (!list_empty(&NFS_I(dir)->open_files)) {
465 nfs_advise_use_readdirplus(dir);
466 nfs_zap_mapping(dir, dir->i_mapping);
471 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
473 struct qstr filename = QSTR_INIT(entry->name, entry->len);
474 struct dentry *dentry;
475 struct dentry *alias;
476 struct inode *dir = parent->d_inode;
480 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
482 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
484 if (filename.name[0] == '.') {
485 if (filename.len == 1)
487 if (filename.len == 2 && filename.name[1] == '.')
490 filename.hash = full_name_hash(filename.name, filename.len);
492 dentry = d_lookup(parent, &filename);
493 if (dentry != NULL) {
494 /* Is there a mountpoint here? If so, just exit */
495 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
496 &entry->fattr->fsid))
498 if (nfs_same_file(dentry, entry)) {
499 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
500 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
502 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
505 d_invalidate(dentry);
510 dentry = d_alloc(parent, &filename);
514 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
518 alias = d_splice_alias(inode, dentry);
522 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
525 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
531 /* Perform conversion from xdr to cache array */
533 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
534 struct page **xdr_pages, struct page *page, unsigned int buflen)
536 struct xdr_stream stream;
538 struct page *scratch;
539 struct nfs_cache_array *array;
540 unsigned int count = 0;
543 scratch = alloc_page(GFP_KERNEL);
547 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
548 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
551 status = xdr_decode(desc, entry, &stream);
553 if (status == -EAGAIN)
561 nfs_prime_dcache(desc->file->f_path.dentry, entry);
563 status = nfs_readdir_add_to_array(entry, page);
566 } while (!entry->eof);
568 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
569 array = nfs_readdir_get_array(page);
570 if (!IS_ERR(array)) {
571 array->eof_index = array->size;
573 nfs_readdir_release_array(page);
575 status = PTR_ERR(array);
583 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
586 for (i = 0; i < npages; i++)
591 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
594 nfs_readdir_free_pagearray(pages, npages);
598 * nfs_readdir_large_page will allocate pages that must be freed with a call
599 * to nfs_readdir_free_large_page
602 int nfs_readdir_large_page(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_pagearray(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 void *pages_ptr = NULL;
624 struct nfs_entry entry;
625 struct file *file = desc->file;
626 struct nfs_cache_array *array;
627 int status = -ENOMEM;
628 unsigned int array_size = ARRAY_SIZE(pages);
630 entry.prev_cookie = 0;
631 entry.cookie = desc->last_cookie;
633 entry.fh = nfs_alloc_fhandle();
634 entry.fattr = nfs_alloc_fattr();
635 entry.server = NFS_SERVER(inode);
636 if (entry.fh == NULL || entry.fattr == NULL)
639 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
640 if (IS_ERR(entry.label)) {
641 status = PTR_ERR(entry.label);
645 array = nfs_readdir_get_array(page);
647 status = PTR_ERR(array);
650 memset(array, 0, sizeof(struct nfs_cache_array));
651 array->eof_index = -1;
653 status = nfs_readdir_large_page(pages, array_size);
655 goto out_release_array;
658 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
663 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
665 if (status == -ENOSPC)
669 } while (array->eof_index < 0);
671 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
673 nfs_readdir_release_array(page);
675 nfs4_label_free(entry.label);
677 nfs_free_fattr(entry.fattr);
678 nfs_free_fhandle(entry.fh);
683 * Now we cache directories properly, by converting xdr information
684 * to an array that can be used for lookups later. This results in
685 * fewer cache pages, since we can store more information on each page.
686 * We only need to convert from xdr once so future lookups are much simpler
689 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
691 struct inode *inode = file_inode(desc->file);
694 ret = nfs_readdir_xdr_to_array(desc, page, inode);
697 SetPageUptodate(page);
699 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
700 /* Should never happen */
701 nfs_zap_mapping(inode, inode->i_mapping);
711 void cache_page_release(nfs_readdir_descriptor_t *desc)
713 if (!desc->page->mapping)
714 nfs_readdir_clear_array(desc->page);
715 page_cache_release(desc->page);
720 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
722 return read_cache_page(file_inode(desc->file)->i_mapping,
723 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
727 * Returns 0 if desc->dir_cookie was found on page desc->page_index
730 int find_cache_page(nfs_readdir_descriptor_t *desc)
734 desc->page = get_cache_page(desc);
735 if (IS_ERR(desc->page))
736 return PTR_ERR(desc->page);
738 res = nfs_readdir_search_array(desc);
740 cache_page_release(desc);
744 /* Search for desc->dir_cookie from the beginning of the page cache */
746 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
750 if (desc->page_index == 0) {
751 desc->current_index = 0;
752 desc->last_cookie = 0;
755 res = find_cache_page(desc);
756 } while (res == -EAGAIN);
761 * Once we've found the start of the dirent within a page: fill 'er up...
764 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
766 struct file *file = desc->file;
769 struct nfs_cache_array *array = NULL;
770 struct nfs_open_dir_context *ctx = file->private_data;
772 array = nfs_readdir_get_array(desc->page);
774 res = PTR_ERR(array);
778 for (i = desc->cache_entry_index; i < array->size; i++) {
779 struct nfs_cache_array_entry *ent;
781 ent = &array->array[i];
782 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
783 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
788 if (i < (array->size-1))
789 *desc->dir_cookie = array->array[i+1].cookie;
791 *desc->dir_cookie = array->last_cookie;
795 if (array->eof_index >= 0)
798 nfs_readdir_release_array(desc->page);
800 cache_page_release(desc);
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 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
851 cache_page_release(desc);
855 static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
857 struct nfs_inode *nfsi = NFS_I(dir);
859 if (nfs_attribute_cache_expired(dir))
861 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
866 /* The file offset position represents the dirent entry number. A
867 last cookie cache takes care of the common case of reading the
870 static int nfs_readdir(struct file *file, struct dir_context *ctx)
872 struct dentry *dentry = file->f_path.dentry;
873 struct inode *inode = dentry->d_inode;
874 nfs_readdir_descriptor_t my_desc,
876 struct nfs_open_dir_context *dir_ctx = file->private_data;
879 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
880 file, (long long)ctx->pos);
881 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
884 * ctx->pos points to the dirent entry number.
885 * *desc->dir_cookie has the cookie for the next entry. We have
886 * to either find the entry with the appropriate number or
887 * revalidate the cookie.
889 memset(desc, 0, sizeof(*desc));
893 desc->dir_cookie = &dir_ctx->dir_cookie;
894 desc->decode = NFS_PROTO(inode)->decode_dirent;
895 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
897 nfs_block_sillyrename(dentry);
898 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
899 res = nfs_revalidate_mapping(inode, file->f_mapping);
904 res = readdir_search_pagecache(desc);
906 if (res == -EBADCOOKIE) {
908 /* This means either end of directory */
909 if (*desc->dir_cookie && desc->eof == 0) {
910 /* Or that the server has 'lost' a cookie */
911 res = uncached_readdir(desc);
917 if (res == -ETOOSMALL && desc->plus) {
918 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
919 nfs_zap_caches(inode);
920 desc->page_index = 0;
928 res = nfs_do_filldir(desc);
931 } while (!desc->eof);
933 nfs_unblock_sillyrename(dentry);
936 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
940 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
942 struct inode *inode = file_inode(filp);
943 struct nfs_open_dir_context *dir_ctx = filp->private_data;
945 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
946 filp, offset, whence);
948 mutex_lock(&inode->i_mutex);
951 offset += filp->f_pos;
959 if (offset != filp->f_pos) {
960 filp->f_pos = offset;
961 dir_ctx->dir_cookie = 0;
965 mutex_unlock(&inode->i_mutex);
970 * All directory operations under NFS are synchronous, so fsync()
971 * is a dummy operation.
973 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
976 struct inode *inode = file_inode(filp);
978 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
980 mutex_lock(&inode->i_mutex);
981 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
982 mutex_unlock(&inode->i_mutex);
987 * nfs_force_lookup_revalidate - Mark the directory as having changed
988 * @dir - pointer to directory inode
990 * This forces the revalidation code in nfs_lookup_revalidate() to do a
991 * full lookup on all child dentries of 'dir' whenever a change occurs
992 * on the server that might have invalidated our dcache.
994 * The caller should be holding dir->i_lock
996 void nfs_force_lookup_revalidate(struct inode *dir)
998 NFS_I(dir)->cache_change_attribute++;
1000 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1003 * A check for whether or not the parent directory has changed.
1004 * In the case it has, we assume that the dentries are untrustworthy
1005 * and may need to be looked up again.
1006 * If rcu_walk prevents us from performing a full check, return 0.
1008 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1013 if (IS_ROOT(dentry))
1015 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1017 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1019 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1021 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir);
1023 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1026 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1032 * Use intent information to check whether or not we're going to do
1033 * an O_EXCL create using this path component.
1035 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1037 if (NFS_PROTO(dir)->version == 2)
1039 return flags & LOOKUP_EXCL;
1043 * Inode and filehandle revalidation for lookups.
1045 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1046 * or if the intent information indicates that we're about to open this
1047 * particular file and the "nocto" mount flag is not set.
1051 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1053 struct nfs_server *server = NFS_SERVER(inode);
1056 if (IS_AUTOMOUNT(inode))
1058 /* VFS wants an on-the-wire revalidation */
1059 if (flags & LOOKUP_REVAL)
1061 /* This is an open(2) */
1062 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1063 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1066 return (inode->i_nlink == 0) ? -ENOENT : 0;
1068 if (flags & LOOKUP_RCU)
1070 ret = __nfs_revalidate_inode(server, inode);
1077 * We judge how long we want to trust negative
1078 * dentries by looking at the parent inode mtime.
1080 * If parent mtime has changed, we revalidate, else we wait for a
1081 * period corresponding to the parent's attribute cache timeout value.
1083 * If LOOKUP_RCU prevents us from performing a full check, return 1
1084 * suggesting a reval is needed.
1087 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1090 /* Don't revalidate a negative dentry if we're creating a new file */
1091 if (flags & LOOKUP_CREATE)
1093 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1095 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1099 * This is called every time the dcache has a lookup hit,
1100 * and we should check whether we can really trust that
1103 * NOTE! The hit can be a negative hit too, don't assume
1106 * If the parent directory is seen to have changed, we throw out the
1107 * cached dentry and do a new lookup.
1109 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1112 struct inode *inode;
1113 struct dentry *parent;
1114 struct nfs_fh *fhandle = NULL;
1115 struct nfs_fattr *fattr = NULL;
1116 struct nfs4_label *label = NULL;
1119 if (flags & LOOKUP_RCU) {
1120 parent = ACCESS_ONCE(dentry->d_parent);
1121 dir = ACCESS_ONCE(parent->d_inode);
1125 parent = dget_parent(dentry);
1126 dir = parent->d_inode;
1128 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1129 inode = dentry->d_inode;
1132 if (nfs_neg_need_reval(dir, dentry, flags)) {
1133 if (flags & LOOKUP_RCU)
1137 goto out_valid_noent;
1140 if (is_bad_inode(inode)) {
1141 if (flags & LOOKUP_RCU)
1143 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1148 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1149 goto out_set_verifier;
1151 /* Force a full look up iff the parent directory has changed */
1152 if (!nfs_is_exclusive_create(dir, flags) &&
1153 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1155 if (nfs_lookup_verify_inode(inode, flags)) {
1156 if (flags & LOOKUP_RCU)
1158 goto out_zap_parent;
1163 if (flags & LOOKUP_RCU)
1166 if (NFS_STALE(inode))
1170 fhandle = nfs_alloc_fhandle();
1171 fattr = nfs_alloc_fattr();
1172 if (fhandle == NULL || fattr == NULL)
1175 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1179 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1180 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1181 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1184 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1186 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1189 nfs_setsecurity(inode, fattr, label);
1191 nfs_free_fattr(fattr);
1192 nfs_free_fhandle(fhandle);
1193 nfs4_label_free(label);
1196 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1198 /* Success: notify readdir to use READDIRPLUS */
1199 nfs_advise_use_readdirplus(dir);
1201 if (flags & LOOKUP_RCU) {
1202 if (parent != ACCESS_ONCE(dentry->d_parent))
1206 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1210 nfs_zap_caches(dir);
1212 WARN_ON(flags & LOOKUP_RCU);
1213 nfs_free_fattr(fattr);
1214 nfs_free_fhandle(fhandle);
1215 nfs4_label_free(label);
1216 nfs_mark_for_revalidate(dir);
1217 if (inode && S_ISDIR(inode->i_mode)) {
1218 /* Purge readdir caches. */
1219 nfs_zap_caches(inode);
1221 * We can't d_drop the root of a disconnected tree:
1222 * its d_hash is on the s_anon list and d_drop() would hide
1223 * it from shrink_dcache_for_unmount(), leading to busy
1224 * inodes on unmount and further oopses.
1226 if (IS_ROOT(dentry))
1230 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1234 WARN_ON(flags & LOOKUP_RCU);
1235 nfs_free_fattr(fattr);
1236 nfs_free_fhandle(fhandle);
1237 nfs4_label_free(label);
1239 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1240 __func__, dentry, error);
1245 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1246 * when we don't really care about the dentry name. This is called when a
1247 * pathwalk ends on a dentry that was not found via a normal lookup in the
1248 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1250 * In this situation, we just want to verify that the inode itself is OK
1251 * since the dentry might have changed on the server.
1253 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1256 struct inode *inode = dentry->d_inode;
1259 * I believe we can only get a negative dentry here in the case of a
1260 * procfs-style symlink. Just assume it's correct for now, but we may
1261 * eventually need to do something more here.
1264 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1269 if (is_bad_inode(inode)) {
1270 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1275 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1276 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1277 __func__, inode->i_ino, error ? "invalid" : "valid");
1282 * This is called from dput() when d_count is going to 0.
1284 static int nfs_dentry_delete(const struct dentry *dentry)
1286 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1287 dentry, dentry->d_flags);
1289 /* Unhash any dentry with a stale inode */
1290 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1293 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1294 /* Unhash it, so that ->d_iput() would be called */
1297 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1298 /* Unhash it, so that ancestors of killed async unlink
1299 * files will be cleaned up during umount */
1306 /* Ensure that we revalidate inode->i_nlink */
1307 static void nfs_drop_nlink(struct inode *inode)
1309 spin_lock(&inode->i_lock);
1310 /* drop the inode if we're reasonably sure this is the last link */
1311 if (inode->i_nlink == 1)
1313 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1314 spin_unlock(&inode->i_lock);
1318 * Called when the dentry loses inode.
1319 * We use it to clean up silly-renamed files.
1321 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1323 if (S_ISDIR(inode->i_mode))
1324 /* drop any readdir cache as it could easily be old */
1325 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1327 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1328 nfs_complete_unlink(dentry, inode);
1329 nfs_drop_nlink(inode);
1334 static void nfs_d_release(struct dentry *dentry)
1336 /* free cached devname value, if it survived that far */
1337 if (unlikely(dentry->d_fsdata)) {
1338 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1341 kfree(dentry->d_fsdata);
1345 const struct dentry_operations nfs_dentry_operations = {
1346 .d_revalidate = nfs_lookup_revalidate,
1347 .d_weak_revalidate = nfs_weak_revalidate,
1348 .d_delete = nfs_dentry_delete,
1349 .d_iput = nfs_dentry_iput,
1350 .d_automount = nfs_d_automount,
1351 .d_release = nfs_d_release,
1353 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1355 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1358 struct dentry *parent;
1359 struct inode *inode = NULL;
1360 struct nfs_fh *fhandle = NULL;
1361 struct nfs_fattr *fattr = NULL;
1362 struct nfs4_label *label = NULL;
1365 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1366 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1368 res = ERR_PTR(-ENAMETOOLONG);
1369 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1373 * If we're doing an exclusive create, optimize away the lookup
1374 * but don't hash the dentry.
1376 if (nfs_is_exclusive_create(dir, flags)) {
1377 d_instantiate(dentry, NULL);
1382 res = ERR_PTR(-ENOMEM);
1383 fhandle = nfs_alloc_fhandle();
1384 fattr = nfs_alloc_fattr();
1385 if (fhandle == NULL || fattr == NULL)
1388 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1392 parent = dentry->d_parent;
1393 /* Protect against concurrent sillydeletes */
1394 trace_nfs_lookup_enter(dir, dentry, flags);
1395 nfs_block_sillyrename(parent);
1396 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1397 if (error == -ENOENT)
1400 res = ERR_PTR(error);
1401 goto out_unblock_sillyrename;
1403 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1404 res = ERR_CAST(inode);
1406 goto out_unblock_sillyrename;
1408 /* Success: notify readdir to use READDIRPLUS */
1409 nfs_advise_use_readdirplus(dir);
1412 res = d_splice_alias(inode, dentry);
1415 goto out_unblock_sillyrename;
1418 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1419 out_unblock_sillyrename:
1420 nfs_unblock_sillyrename(parent);
1421 trace_nfs_lookup_exit(dir, dentry, flags, error);
1422 nfs4_label_free(label);
1424 nfs_free_fattr(fattr);
1425 nfs_free_fhandle(fhandle);
1428 EXPORT_SYMBOL_GPL(nfs_lookup);
1430 #if IS_ENABLED(CONFIG_NFS_V4)
1431 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1433 const struct dentry_operations nfs4_dentry_operations = {
1434 .d_revalidate = nfs4_lookup_revalidate,
1435 .d_delete = nfs_dentry_delete,
1436 .d_iput = nfs_dentry_iput,
1437 .d_automount = nfs_d_automount,
1438 .d_release = nfs_d_release,
1440 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1442 static fmode_t flags_to_mode(int flags)
1444 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1445 if ((flags & O_ACCMODE) != O_WRONLY)
1447 if ((flags & O_ACCMODE) != O_RDONLY)
1452 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1454 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1457 static int do_open(struct inode *inode, struct file *filp)
1459 nfs_fscache_open_file(inode, filp);
1463 static int nfs_finish_open(struct nfs_open_context *ctx,
1464 struct dentry *dentry,
1465 struct file *file, unsigned open_flags,
1470 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1471 *opened |= FILE_CREATED;
1473 err = finish_open(file, dentry, do_open, opened);
1476 nfs_file_set_open_context(file, ctx);
1482 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1483 struct file *file, unsigned open_flags,
1484 umode_t mode, int *opened)
1486 struct nfs_open_context *ctx;
1488 struct iattr attr = { .ia_valid = ATTR_OPEN };
1489 struct inode *inode;
1490 unsigned int lookup_flags = 0;
1493 /* Expect a negative dentry */
1494 BUG_ON(dentry->d_inode);
1496 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1497 dir->i_sb->s_id, dir->i_ino, dentry);
1499 err = nfs_check_flags(open_flags);
1503 /* NFS only supports OPEN on regular files */
1504 if ((open_flags & O_DIRECTORY)) {
1505 if (!d_unhashed(dentry)) {
1507 * Hashed negative dentry with O_DIRECTORY: dentry was
1508 * revalidated and is fine, no need to perform lookup
1513 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1517 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1518 return -ENAMETOOLONG;
1520 if (open_flags & O_CREAT) {
1521 attr.ia_valid |= ATTR_MODE;
1522 attr.ia_mode = mode & ~current_umask();
1524 if (open_flags & O_TRUNC) {
1525 attr.ia_valid |= ATTR_SIZE;
1529 ctx = create_nfs_open_context(dentry, open_flags);
1534 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1535 nfs_block_sillyrename(dentry->d_parent);
1536 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1537 nfs_unblock_sillyrename(dentry->d_parent);
1538 if (IS_ERR(inode)) {
1539 err = PTR_ERR(inode);
1540 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1541 put_nfs_open_context(ctx);
1545 d_add(dentry, NULL);
1546 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1552 if (!(open_flags & O_NOFOLLOW))
1562 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1563 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1564 put_nfs_open_context(ctx);
1569 res = nfs_lookup(dir, dentry, lookup_flags);
1574 return finish_no_open(file, res);
1576 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1578 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1580 struct inode *inode;
1583 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1585 if (d_mountpoint(dentry))
1587 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1590 inode = dentry->d_inode;
1592 /* We can't create new files in nfs_open_revalidate(), so we
1593 * optimize away revalidation of negative dentries.
1595 if (inode == NULL) {
1596 struct dentry *parent;
1599 if (flags & LOOKUP_RCU) {
1600 parent = ACCESS_ONCE(dentry->d_parent);
1601 dir = ACCESS_ONCE(parent->d_inode);
1605 parent = dget_parent(dentry);
1606 dir = parent->d_inode;
1608 if (!nfs_neg_need_reval(dir, dentry, flags))
1610 else if (flags & LOOKUP_RCU)
1612 if (!(flags & LOOKUP_RCU))
1614 else if (parent != ACCESS_ONCE(dentry->d_parent))
1619 /* NFS only supports OPEN on regular files */
1620 if (!S_ISREG(inode->i_mode))
1622 /* We cannot do exclusive creation on a positive dentry */
1623 if (flags & LOOKUP_EXCL)
1626 /* Let f_op->open() actually open (and revalidate) the file */
1633 return nfs_lookup_revalidate(dentry, flags);
1636 #endif /* CONFIG_NFSV4 */
1639 * Code common to create, mkdir, and mknod.
1641 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1642 struct nfs_fattr *fattr,
1643 struct nfs4_label *label)
1645 struct dentry *parent = dget_parent(dentry);
1646 struct inode *dir = parent->d_inode;
1647 struct inode *inode;
1648 int error = -EACCES;
1652 /* We may have been initialized further down */
1653 if (dentry->d_inode)
1655 if (fhandle->size == 0) {
1656 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1660 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1661 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1662 struct nfs_server *server = NFS_SB(dentry->d_sb);
1663 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1667 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1668 error = PTR_ERR(inode);
1671 d_add(dentry, inode);
1676 nfs_mark_for_revalidate(dir);
1680 EXPORT_SYMBOL_GPL(nfs_instantiate);
1683 * Following a failed create operation, we drop the dentry rather
1684 * than retain a negative dentry. This avoids a problem in the event
1685 * that the operation succeeded on the server, but an error in the
1686 * reply path made it appear to have failed.
1688 int nfs_create(struct inode *dir, struct dentry *dentry,
1689 umode_t mode, bool excl)
1692 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1695 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1696 dir->i_sb->s_id, dir->i_ino, dentry);
1698 attr.ia_mode = mode;
1699 attr.ia_valid = ATTR_MODE;
1701 trace_nfs_create_enter(dir, dentry, open_flags);
1702 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1703 trace_nfs_create_exit(dir, dentry, open_flags, error);
1711 EXPORT_SYMBOL_GPL(nfs_create);
1714 * See comments for nfs_proc_create regarding failed operations.
1717 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1722 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1723 dir->i_sb->s_id, dir->i_ino, dentry);
1725 if (!new_valid_dev(rdev))
1728 attr.ia_mode = mode;
1729 attr.ia_valid = ATTR_MODE;
1731 trace_nfs_mknod_enter(dir, dentry);
1732 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1733 trace_nfs_mknod_exit(dir, dentry, status);
1741 EXPORT_SYMBOL_GPL(nfs_mknod);
1744 * See comments for nfs_proc_create regarding failed operations.
1746 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1751 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1752 dir->i_sb->s_id, dir->i_ino, dentry);
1754 attr.ia_valid = ATTR_MODE;
1755 attr.ia_mode = mode | S_IFDIR;
1757 trace_nfs_mkdir_enter(dir, dentry);
1758 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1759 trace_nfs_mkdir_exit(dir, dentry, error);
1767 EXPORT_SYMBOL_GPL(nfs_mkdir);
1769 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1771 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1775 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1779 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1780 dir->i_sb->s_id, dir->i_ino, dentry);
1782 trace_nfs_rmdir_enter(dir, dentry);
1783 if (dentry->d_inode) {
1784 nfs_wait_on_sillyrename(dentry);
1785 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1786 /* Ensure the VFS deletes this inode */
1789 clear_nlink(dentry->d_inode);
1792 nfs_dentry_handle_enoent(dentry);
1795 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1796 trace_nfs_rmdir_exit(dir, dentry, error);
1800 EXPORT_SYMBOL_GPL(nfs_rmdir);
1803 * Remove a file after making sure there are no pending writes,
1804 * and after checking that the file has only one user.
1806 * We invalidate the attribute cache and free the inode prior to the operation
1807 * to avoid possible races if the server reuses the inode.
1809 static int nfs_safe_remove(struct dentry *dentry)
1811 struct inode *dir = dentry->d_parent->d_inode;
1812 struct inode *inode = dentry->d_inode;
1815 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1817 /* If the dentry was sillyrenamed, we simply call d_delete() */
1818 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1823 trace_nfs_remove_enter(dir, dentry);
1824 if (inode != NULL) {
1825 NFS_PROTO(inode)->return_delegation(inode);
1826 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1828 nfs_drop_nlink(inode);
1830 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1831 if (error == -ENOENT)
1832 nfs_dentry_handle_enoent(dentry);
1833 trace_nfs_remove_exit(dir, dentry, error);
1838 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1839 * belongs to an active ".nfs..." file and we return -EBUSY.
1841 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1843 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1846 int need_rehash = 0;
1848 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1849 dir->i_ino, dentry);
1851 trace_nfs_unlink_enter(dir, dentry);
1852 spin_lock(&dentry->d_lock);
1853 if (d_count(dentry) > 1) {
1854 spin_unlock(&dentry->d_lock);
1855 /* Start asynchronous writeout of the inode */
1856 write_inode_now(dentry->d_inode, 0);
1857 error = nfs_sillyrename(dir, dentry);
1860 if (!d_unhashed(dentry)) {
1864 spin_unlock(&dentry->d_lock);
1865 error = nfs_safe_remove(dentry);
1866 if (!error || error == -ENOENT) {
1867 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1868 } else if (need_rehash)
1871 trace_nfs_unlink_exit(dir, dentry, error);
1874 EXPORT_SYMBOL_GPL(nfs_unlink);
1877 * To create a symbolic link, most file systems instantiate a new inode,
1878 * add a page to it containing the path, then write it out to the disk
1879 * using prepare_write/commit_write.
1881 * Unfortunately the NFS client can't create the in-core inode first
1882 * because it needs a file handle to create an in-core inode (see
1883 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1884 * symlink request has completed on the server.
1886 * So instead we allocate a raw page, copy the symname into it, then do
1887 * the SYMLINK request with the page as the buffer. If it succeeds, we
1888 * now have a new file handle and can instantiate an in-core NFS inode
1889 * and move the raw page into its mapping.
1891 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1896 unsigned int pathlen = strlen(symname);
1899 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1900 dir->i_ino, dentry, symname);
1902 if (pathlen > PAGE_SIZE)
1903 return -ENAMETOOLONG;
1905 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1906 attr.ia_valid = ATTR_MODE;
1908 page = alloc_page(GFP_HIGHUSER);
1912 kaddr = kmap_atomic(page);
1913 memcpy(kaddr, symname, pathlen);
1914 if (pathlen < PAGE_SIZE)
1915 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1916 kunmap_atomic(kaddr);
1918 trace_nfs_symlink_enter(dir, dentry);
1919 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1920 trace_nfs_symlink_exit(dir, dentry, error);
1922 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1923 dir->i_sb->s_id, dir->i_ino,
1924 dentry, symname, error);
1931 * No big deal if we can't add this page to the page cache here.
1932 * READLINK will get the missing page from the server if needed.
1934 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1936 SetPageUptodate(page);
1939 * add_to_page_cache_lru() grabs an extra page refcount.
1940 * Drop it here to avoid leaking this page later.
1942 page_cache_release(page);
1948 EXPORT_SYMBOL_GPL(nfs_symlink);
1951 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1953 struct inode *inode = old_dentry->d_inode;
1956 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1957 old_dentry, dentry);
1959 trace_nfs_link_enter(inode, dir, dentry);
1960 NFS_PROTO(inode)->return_delegation(inode);
1963 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1966 d_add(dentry, inode);
1968 trace_nfs_link_exit(inode, dir, dentry, error);
1971 EXPORT_SYMBOL_GPL(nfs_link);
1975 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1976 * different file handle for the same inode after a rename (e.g. when
1977 * moving to a different directory). A fail-safe method to do so would
1978 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1979 * rename the old file using the sillyrename stuff. This way, the original
1980 * file in old_dir will go away when the last process iput()s the inode.
1984 * It actually works quite well. One needs to have the possibility for
1985 * at least one ".nfs..." file in each directory the file ever gets
1986 * moved or linked to which happens automagically with the new
1987 * implementation that only depends on the dcache stuff instead of
1988 * using the inode layer
1990 * Unfortunately, things are a little more complicated than indicated
1991 * above. For a cross-directory move, we want to make sure we can get
1992 * rid of the old inode after the operation. This means there must be
1993 * no pending writes (if it's a file), and the use count must be 1.
1994 * If these conditions are met, we can drop the dentries before doing
1997 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1998 struct inode *new_dir, struct dentry *new_dentry)
2000 struct inode *old_inode = old_dentry->d_inode;
2001 struct inode *new_inode = new_dentry->d_inode;
2002 struct dentry *dentry = NULL, *rehash = NULL;
2003 struct rpc_task *task;
2006 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2007 old_dentry, new_dentry,
2008 d_count(new_dentry));
2010 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2012 * For non-directories, check whether the target is busy and if so,
2013 * make a copy of the dentry and then do a silly-rename. If the
2014 * silly-rename succeeds, the copied dentry is hashed and becomes
2017 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2019 * To prevent any new references to the target during the
2020 * rename, we unhash the dentry in advance.
2022 if (!d_unhashed(new_dentry)) {
2024 rehash = new_dentry;
2027 if (d_count(new_dentry) > 2) {
2030 /* copy the target dentry's name */
2031 dentry = d_alloc(new_dentry->d_parent,
2032 &new_dentry->d_name);
2036 /* silly-rename the existing target ... */
2037 err = nfs_sillyrename(new_dir, new_dentry);
2041 new_dentry = dentry;
2047 NFS_PROTO(old_inode)->return_delegation(old_inode);
2048 if (new_inode != NULL)
2049 NFS_PROTO(new_inode)->return_delegation(new_inode);
2051 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2053 error = PTR_ERR(task);
2057 error = rpc_wait_for_completion_task(task);
2059 error = task->tk_status;
2061 nfs_mark_for_revalidate(old_inode);
2065 trace_nfs_rename_exit(old_dir, old_dentry,
2066 new_dir, new_dentry, error);
2068 if (new_inode != NULL)
2069 nfs_drop_nlink(new_inode);
2070 d_move(old_dentry, new_dentry);
2071 nfs_set_verifier(new_dentry,
2072 nfs_save_change_attribute(new_dir));
2073 } else if (error == -ENOENT)
2074 nfs_dentry_handle_enoent(old_dentry);
2076 /* new dentry created? */
2081 EXPORT_SYMBOL_GPL(nfs_rename);
2083 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2084 static LIST_HEAD(nfs_access_lru_list);
2085 static atomic_long_t nfs_access_nr_entries;
2087 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2088 module_param(nfs_access_max_cachesize, ulong, 0644);
2089 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2091 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2093 put_rpccred(entry->cred);
2094 kfree_rcu(entry, rcu_head);
2095 smp_mb__before_atomic();
2096 atomic_long_dec(&nfs_access_nr_entries);
2097 smp_mb__after_atomic();
2100 static void nfs_access_free_list(struct list_head *head)
2102 struct nfs_access_entry *cache;
2104 while (!list_empty(head)) {
2105 cache = list_entry(head->next, struct nfs_access_entry, lru);
2106 list_del(&cache->lru);
2107 nfs_access_free_entry(cache);
2111 static unsigned long
2112 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2115 struct nfs_inode *nfsi, *next;
2116 struct nfs_access_entry *cache;
2119 spin_lock(&nfs_access_lru_lock);
2120 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2121 struct inode *inode;
2123 if (nr_to_scan-- == 0)
2125 inode = &nfsi->vfs_inode;
2126 spin_lock(&inode->i_lock);
2127 if (list_empty(&nfsi->access_cache_entry_lru))
2128 goto remove_lru_entry;
2129 cache = list_entry(nfsi->access_cache_entry_lru.next,
2130 struct nfs_access_entry, lru);
2131 list_move(&cache->lru, &head);
2132 rb_erase(&cache->rb_node, &nfsi->access_cache);
2134 if (!list_empty(&nfsi->access_cache_entry_lru))
2135 list_move_tail(&nfsi->access_cache_inode_lru,
2136 &nfs_access_lru_list);
2139 list_del_init(&nfsi->access_cache_inode_lru);
2140 smp_mb__before_atomic();
2141 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2142 smp_mb__after_atomic();
2144 spin_unlock(&inode->i_lock);
2146 spin_unlock(&nfs_access_lru_lock);
2147 nfs_access_free_list(&head);
2152 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2154 int nr_to_scan = sc->nr_to_scan;
2155 gfp_t gfp_mask = sc->gfp_mask;
2157 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2159 return nfs_do_access_cache_scan(nr_to_scan);
2164 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2166 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2170 nfs_access_cache_enforce_limit(void)
2172 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2174 unsigned int nr_to_scan;
2176 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2179 diff = nr_entries - nfs_access_max_cachesize;
2180 if (diff < nr_to_scan)
2182 nfs_do_access_cache_scan(nr_to_scan);
2185 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2187 struct rb_root *root_node = &nfsi->access_cache;
2189 struct nfs_access_entry *entry;
2191 /* Unhook entries from the cache */
2192 while ((n = rb_first(root_node)) != NULL) {
2193 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2194 rb_erase(n, root_node);
2195 list_move(&entry->lru, head);
2197 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2200 void nfs_access_zap_cache(struct inode *inode)
2204 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2206 /* Remove from global LRU init */
2207 spin_lock(&nfs_access_lru_lock);
2208 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2209 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2211 spin_lock(&inode->i_lock);
2212 __nfs_access_zap_cache(NFS_I(inode), &head);
2213 spin_unlock(&inode->i_lock);
2214 spin_unlock(&nfs_access_lru_lock);
2215 nfs_access_free_list(&head);
2217 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2219 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2221 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2222 struct nfs_access_entry *entry;
2225 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2227 if (cred < entry->cred)
2229 else if (cred > entry->cred)
2237 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2239 struct nfs_inode *nfsi = NFS_I(inode);
2240 struct nfs_access_entry *cache;
2243 spin_lock(&inode->i_lock);
2244 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2246 cache = nfs_access_search_rbtree(inode, cred);
2249 if (!nfs_have_delegated_attributes(inode) &&
2250 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2252 res->jiffies = cache->jiffies;
2253 res->cred = cache->cred;
2254 res->mask = cache->mask;
2255 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2258 spin_unlock(&inode->i_lock);
2261 rb_erase(&cache->rb_node, &nfsi->access_cache);
2262 list_del(&cache->lru);
2263 spin_unlock(&inode->i_lock);
2264 nfs_access_free_entry(cache);
2267 spin_unlock(&inode->i_lock);
2268 nfs_access_zap_cache(inode);
2272 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2274 /* Only check the most recently returned cache entry,
2275 * but do it without locking.
2277 struct nfs_inode *nfsi = NFS_I(inode);
2278 struct nfs_access_entry *cache;
2280 struct list_head *lh;
2283 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2285 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2286 cache = list_entry(lh, struct nfs_access_entry, lru);
2287 if (lh == &nfsi->access_cache_entry_lru ||
2288 cred != cache->cred)
2292 if (!nfs_have_delegated_attributes(inode) &&
2293 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2295 res->jiffies = cache->jiffies;
2296 res->cred = cache->cred;
2297 res->mask = cache->mask;
2304 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2306 struct nfs_inode *nfsi = NFS_I(inode);
2307 struct rb_root *root_node = &nfsi->access_cache;
2308 struct rb_node **p = &root_node->rb_node;
2309 struct rb_node *parent = NULL;
2310 struct nfs_access_entry *entry;
2312 spin_lock(&inode->i_lock);
2313 while (*p != NULL) {
2315 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2317 if (set->cred < entry->cred)
2318 p = &parent->rb_left;
2319 else if (set->cred > entry->cred)
2320 p = &parent->rb_right;
2324 rb_link_node(&set->rb_node, parent, p);
2325 rb_insert_color(&set->rb_node, root_node);
2326 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2327 spin_unlock(&inode->i_lock);
2330 rb_replace_node(parent, &set->rb_node, root_node);
2331 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2332 list_del(&entry->lru);
2333 spin_unlock(&inode->i_lock);
2334 nfs_access_free_entry(entry);
2337 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2339 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2342 RB_CLEAR_NODE(&cache->rb_node);
2343 cache->jiffies = set->jiffies;
2344 cache->cred = get_rpccred(set->cred);
2345 cache->mask = set->mask;
2347 /* The above field assignments must be visible
2348 * before this item appears on the lru. We cannot easily
2349 * use rcu_assign_pointer, so just force the memory barrier.
2352 nfs_access_add_rbtree(inode, cache);
2354 /* Update accounting */
2355 smp_mb__before_atomic();
2356 atomic_long_inc(&nfs_access_nr_entries);
2357 smp_mb__after_atomic();
2359 /* Add inode to global LRU list */
2360 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2361 spin_lock(&nfs_access_lru_lock);
2362 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2363 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2364 &nfs_access_lru_list);
2365 spin_unlock(&nfs_access_lru_lock);
2367 nfs_access_cache_enforce_limit();
2369 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2371 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2374 if (access_result & NFS4_ACCESS_READ)
2375 entry->mask |= MAY_READ;
2377 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2378 entry->mask |= MAY_WRITE;
2379 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2380 entry->mask |= MAY_EXEC;
2382 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2384 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2386 struct nfs_access_entry cache;
2389 trace_nfs_access_enter(inode);
2391 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2393 status = nfs_access_get_cached(inode, cred, &cache);
2398 if (mask & MAY_NOT_BLOCK)
2401 /* Be clever: ask server to check for all possible rights */
2402 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2404 cache.jiffies = jiffies;
2405 status = NFS_PROTO(inode)->access(inode, &cache);
2407 if (status == -ESTALE) {
2408 nfs_zap_caches(inode);
2409 if (!S_ISDIR(inode->i_mode))
2410 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2414 nfs_access_add_cache(inode, &cache);
2416 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2419 trace_nfs_access_exit(inode, status);
2423 static int nfs_open_permission_mask(int openflags)
2427 if (openflags & __FMODE_EXEC) {
2428 /* ONLY check exec rights */
2431 if ((openflags & O_ACCMODE) != O_WRONLY)
2433 if ((openflags & O_ACCMODE) != O_RDONLY)
2440 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2442 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2444 EXPORT_SYMBOL_GPL(nfs_may_open);
2446 int nfs_permission(struct inode *inode, int mask)
2448 struct rpc_cred *cred;
2451 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2453 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2455 /* Is this sys_access() ? */
2456 if (mask & (MAY_ACCESS | MAY_CHDIR))
2459 switch (inode->i_mode & S_IFMT) {
2466 * Optimize away all write operations, since the server
2467 * will check permissions when we perform the op.
2469 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2474 if (!NFS_PROTO(inode)->access)
2477 /* Always try fast lookups first */
2479 cred = rpc_lookup_cred_nonblock();
2481 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2483 res = PTR_ERR(cred);
2485 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2486 /* Fast lookup failed, try the slow way */
2487 cred = rpc_lookup_cred();
2488 if (!IS_ERR(cred)) {
2489 res = nfs_do_access(inode, cred, mask);
2492 res = PTR_ERR(cred);
2495 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2498 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2499 inode->i_sb->s_id, inode->i_ino, mask, res);
2502 if (mask & MAY_NOT_BLOCK)
2505 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2507 res = generic_permission(inode, mask);
2510 EXPORT_SYMBOL_GPL(nfs_permission);
2514 * version-control: t
2515 * kept-new-versions: 5